From wade@cs.utk.edu Tue Feb 9 10:47:32 1993 Received: from THUD.CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA26386; Tue, 9 Feb 93 10:47:32 -0500 Received: from LOCALHOST.cs.utk.edu by thud.cs.utk.edu with SMTP (5.61++/2.7c-UTK) id AA26430; Tue, 9 Feb 93 10:47:26 -0500 Message-Id: <9302091547.AA26430@thud.cs.utk.edu> To: pbwg-comm-archive@surfer.EPM.ORNL.GOV Cc: wade@cs.utk.edu Subject: test Date: Tue, 09 Feb 93 10:47:25 EST From: Reed Wade From owner-pbwg-comm@CS.UTK.EDU Sun Feb 14 16:56:44 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA14958; Sun, 14 Feb 93 16:56:44 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA24403; Sun, 14 Feb 93 16:56:04 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Sun, 14 Feb 1993 16:56:03 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from DASHER.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA24397; Sun, 14 Feb 93 16:56:02 -0500 From: Jack Dongarra Received: by dasher.cs.utk.edu (5.61++/2.7c-UTK) id AA01941; Sun, 14 Feb 93 16:56:01 -0500 Date: Sun, 14 Feb 93 16:56:01 -0500 Message-Id: <9302142156.AA01941@dasher.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: Parallel Benchmark Working Group Meeting For planning purposes, I would like to know how many people will be attending the Parallel Benchmark Working Group (PBWG) meeting on March 1th and 2nd, 1993 in Knoxville, Tennessee. Let me know if your will or will not be attending the meeting. Best wishes, Jack From @ecs.soton.ac.uk,@diana.ecs.soton.ac.uk:C.D.Collier@ecs.southampton.ac.uk Mon Jan 11 12:30:42 1993 Return-Path: <@ecs.soton.ac.uk,@diana.ecs.soton.ac.uk:C.D.Collier@ecs.southampton.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA04153; Mon, 11 Jan 93 12:30:09 -0500 Via: uk.ac.southampton.ecs; Mon, 11 Jan 1993 17:27:19 +0000 Via: diana.ecs.soton.ac.uk; Mon, 11 Jan 93 17:20:06 GMT From: Christine Collier Message-Id: <6780.9301111725@diana.ecs.soton.ac.uk> Subject: Meeting Minneapolis 18th November, 1992 To: dbailey@nas.nasa.com, iyb@lanl.gov, sbveit@ksv.com, carterl@watson.ibm.com, thec@newton.national-physical-lab.co.uk, dongarra@cs.utk.edu, dem@cxa.dl.ac.uk, j.flemming@cray.com, gcfe@npac.syr.edu, danielf@kgnvma.vnet.ibm.com, paulg@meiko.com, gent@genias.de, cmg@cray.com, harp@revmes.mod.uk, siamak@fai.com, hempel@gmd.de, ajgh@ecs.soton.ac.uk, rwh@uk.pac.soton.ecs, j1mart@kgnvmz.vnet.ibm.com, hcooke@parsys.co.uk, messina@caltech.edu, Bminto@cray.com, dennisp@think.com, schneid@csrd.uiuc.edu, simon@nas.nasa.gov, actstea@ml.ruu.cc, frannie@Parsytech.de Date: Mon, 11 Jan 1993 17:25:46 +0000 (GMT) X-Mailer: ELM [version 2.4 PL8] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 7692 Status: RO Parallel Benchmark Working Group Summary of 1st Meeting held in Minneapolis Convention Centre Wednesday November 18th, 1992 1. Introduction and Welcome The meeting was opened by Professor Roger Hockney who welcomed all the participants and asked Tony Hey to say a few words about the background to the meeting. Tony Hey outlined the history of attempts in Europe to establish credible and useful benchmarks for the evaluation of Distributed Memory MIMD systems. At the time the European Genesis work began in 1988 there were no suitable DM message-passing benchmarks - although the Caltech group also undertook a study at about the same time. Four years later the scene has moved on, and besides the Genesis benchmarks for message-passing Fortran programs, HPF/Fortran 90 benchmarks are now appearing. The Perfect Club are also now looking at benchmarks for DM systems and the NASA Ames 'Pencil and Paper' benchmarks are being taken seriously by the vendors. From a situation where there were very few suitable benchmark codes we are now approaching a situation where there is an over-abundance of benchmarks and inevitable duplication of effort. This is undesirable both for the vendors and for the end users of systems. With the adoption of HPF and the current proposal for a standard message-passing interface (MPI), discussed yesterday at the message-passing standards workshop, there seems to be a real window of opportunity to gather together US and European benchmarkers to agree on a useful subset. Roger Hockney then asked those present to introduce themselves and make some remark about their interest in this activity. A list of attendees is attached to this summary. There was widespread support for such an activity and a summary of contributions as follows: Horst Simon of NASA Ames suggested that we follow the HPF model and form working groups in identified areas. He also voiced reservations about any attempt to create a fully comprehensive benchmark suite. Denis Parkinson of TMC supported the activity but pointed out that it was important that this did not become "yet another" set of benchmarks for vendors to implement. He also stressed the need for HPF versions as well as message-passing versions, and raised the question of scalability of benchmarks. Aard van der Steen raised the question of Japanese participation in such an activity. Siamak Hassanzadeh of Fujitsu supported the proposal and suggested the inclusion of seismic benchmarking codes. Rolf Hempel of the GMD in Sankt Augustin stressed his organization's support for the activity and briefly described the RAPS benchmarking initiative in Europe. This is a consortium of users and software houses, led by ECMWF and including ESI and AVL, supported by Convex, Cray, Fujitsu, IBM, Intel and Meiko RAPS is an acronym for Real Applications on Parallel Systems. The Genesis benchmark codes are included as a public domain subset of RAPS but the major codes would not be publicly available. Bill Minto of Cray UK stressed that real applications were important and that it was also necessary to address as wide a spread of applications as possible. Geoffrey Fox of Syracuse suggested that it was only realistic at this stage at attempt to co-ordinate benchmarking activities. His activities at Syracuse had lately been concerned with constructing benchmarks for validating HPF and Fortran 90 compilers. Trevor Chambers of NPL, UK gave his support to the activity and announced that a major new European benchmarking project PEPS with NPL were involvement has just started. Gordon Harp of DRA, Malvern UK said that the DRA (Defence Research Agency) were involved in a collaboration with agencies in the US, Canada and Australia looking at benchmarks for defence applications. He was concerned that there was a need for real applications and issues such as scalability and power consumption were important from the DRA perspective. Francis Wray of Parsytec and Paul Garrett of Meiko both voiced concern that this activity should not generate more rather than less work in procurement activities. David Schneider from CRSD Illinois, representing the Perfect Club, welcomed an attempt to eliminate redundant effort. He stressed that benchmarks had multiple uses - such as education and compiler evaluation - as well as specific application knowledge. He thought that there may be budget problems if a very organised activity was envisaged. Benchmarks spanning a range of architectures and the need for public domain codes were also stressed. Joanne Martin of the HPSSL IBM Kingston welcomed the initiative and stressed how important it was for vendors not to be confronted with many sets of "standard" benchmarks. 2. Objectives There were no objections to the draft objectives for the group. These objectives are: 1. To establish a comprehensive set of parallel benchmarks that is generally accepted by both users and vendors of parallel system. 2. To provide a focus for parallel benchmark activities and avoid unnecessary duplication of effort and proliferation of benchmarks. 3. To set standards for benchmarking methodology and result-reporting together with a control database/repository for both the benchmarks and the results. 3. Mode of Working 3.1 It was agreed that on HPF-like forum style of working should be adopted with a view to convergence to an agreed set of benchmarks and procedures within 12 months. 3.2 There was not seen to be a need for meetings every six weeks but in order to generate momentum for the project it was thought that two meetings a year were too few. 3.3 Jack Dongarra volunteered to set up a database for benchmarks and results at ORNL. NPL were willing to maintain a European copy of this database. 3.4 It was agreed that all present should send existing benchmarks to Jack Dongarra at ORNL. 3.5 Jack Dongarra and Aard van der Steen agreed to examine the available benchmarks submitted and attempt to classify them appropriately. 3.6 Three other working groups were identified with named individuals taking responsibility for each group. These were as follows: Methodology: Bailey, Hockney and Schneider Kernel Benchmarks: Dongarra, Hockney, van der Steen, Wray Compiler Benchmarks: Fox, Grassl 3.7 A number of application areas were discussed as possible working groups e.g. CFD, Seismic, QCD etc., but it was thought premature to activate such groups at this time. 4. Future Activity 4.1 Jack Dongarra agreed to set up a mail refector at ORNL for the Parallel Benchmark Working Group (PBWG) and to organize the relevant subgroups along the lines of the HPF forum. 4.2 Jack Dongarra also agreed to host the next meeting of the PBWG. Subsequent discussions after the formal close of the meeting led to the dates of March 1st/2nd being selected. 4.3 At the March meeting, each subgroup will produce a discussion document and a benchmark database classification will be proposed. Further discussion on procedures for the PBWG was also deferred until then. From owner-pbwg-comm@CS.UTK.EDU Sun Feb 21 12:08:07 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA10794; Sun, 21 Feb 93 12:08:07 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA19152; Sun, 21 Feb 93 12:07:34 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Sun, 21 Feb 1993 12:07:32 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from THUD.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA19145; Sun, 21 Feb 93 12:07:28 -0500 From: Jack Dongarra Received: by thud.cs.utk.edu (5.61++/2.7c-UTK) id AA05996; Sun, 21 Feb 93 12:07:24 -0500 Date: Sun, 21 Feb 93 12:07:24 -0500 Message-Id: <9302211707.AA05996@thud.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: change of room for the PBWG meeting We have had a change in the meeting room for the Parallel Benchmark Working Group. The new meeting room is in the University Center Room 201. The postscript file below contains a map that may help. Look forward to seeing you next week. Regards, Jack %!PS-Adobe-2.0 EPSF-1.2 %%DocumentFonts: Helvetica-Bold Courier Courier-Bold Times-Bold %%Pages: 1 %%BoundingBox: 39 -113 604 767 %%EndComments /arrowHeight 10 def /arrowWidth 5 def /IdrawDict 54 dict def IdrawDict begin /reencodeISO { dup dup findfont dup length dict begin { 1 index /FID ne { def }{ pop pop } ifelse } forall /Encoding ISOLatin1Encoding def currentdict end definefont } def /ISOLatin1Encoding [ /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright /parenleft/parenright/asterisk/plus/comma/minus/period/slash /zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon /less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N /O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright /asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m /n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/dotlessi/grave/acute/circumflex/tilde/macron/breve /dotaccent/dieresis/.notdef/ring/cedilla/.notdef/hungarumlaut /ogonek/caron/space/exclamdown/cent/sterling/currency/yen/brokenbar /section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot /hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior /acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine /guillemotright/onequarter/onehalf/threequarters/questiondown /Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla /Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex /Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis /multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute /Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis /aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave /iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex /otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis /yacute/thorn/ydieresis ] def /Helvetica-Bold reencodeISO def /Courier reencodeISO def /Courier-Bold reencodeISO def /Times-Bold reencodeISO def /none null def /numGraphicParameters 17 def /stringLimit 65535 def /Begin { save numGraphicParameters dict begin } def /End { end restore } def /SetB { dup type /nulltype eq { pop false /brushRightArrow idef false /brushLeftArrow idef true /brushNone idef } { /brushDashOffset idef /brushDashArray idef 0 ne /brushRightArrow idef 0 ne /brushLeftArrow idef /brushWidth idef false /brushNone idef } ifelse } def /SetCFg { /fgblue idef /fggreen idef /fgred idef } def /SetCBg { /bgblue idef /bggreen idef /bgred idef } def /SetF { /printSize idef /printFont idef } def /SetP { dup type /nulltype eq { pop true /patternNone idef } { dup -1 eq { /patternGrayLevel idef /patternString idef } { /patternGrayLevel idef } ifelse false /patternNone idef } ifelse } def /BSpl { 0 begin storexyn newpath n 1 gt { 0 0 0 0 0 0 1 1 true subspline n 2 gt { 0 0 0 0 1 1 2 2 false subspline 1 1 n 3 sub { /i exch def i 1 sub dup i dup i 1 add dup i 2 add dup false subspline } for n 3 sub dup n 2 sub dup n 1 sub dup 2 copy false subspline } if n 2 sub dup n 1 sub dup 2 copy 2 copy false subspline patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if brushNone not { istroke } if 0 0 1 1 leftarrow n 2 sub dup n 1 sub dup rightarrow } if end } dup 0 4 dict put def /Circ { newpath 0 360 arc patternNone not { ifill } if brushNone not { istroke } if } def /CBSpl { 0 begin dup 2 gt { storexyn newpath n 1 sub dup 0 0 1 1 2 2 true subspline 1 1 n 3 sub { /i exch def i 1 sub dup i dup i 1 add dup i 2 add dup false subspline } for n 3 sub dup n 2 sub dup n 1 sub dup 0 0 false subspline n 2 sub dup n 1 sub dup 0 0 1 1 false subspline patternNone not { ifill } if brushNone not { istroke } if } { Poly } ifelse end } dup 0 4 dict put def /Elli { 0 begin newpath 4 2 roll translate scale 0 0 1 0 360 arc patternNone not { ifill } if brushNone not { istroke } if end } dup 0 1 dict put def /Line { 0 begin 2 storexyn newpath x 0 get y 0 get moveto x 1 get y 1 get lineto brushNone not { istroke } if 0 0 1 1 leftarrow 0 0 1 1 rightarrow end } dup 0 4 dict put def /MLine { 0 begin storexyn newpath n 1 gt { x 0 get y 0 get moveto 1 1 n 1 sub { /i exch def x i get y i get lineto } for patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if brushNone not { istroke } if 0 0 1 1 leftarrow n 2 sub dup n 1 sub dup rightarrow } if end } dup 0 4 dict put def /Poly { 3 1 roll newpath moveto -1 add { lineto } repeat closepath patternNone not { ifill } if brushNone not { istroke } if } def /Rect { 0 begin /t exch def /r exch def /b exch def /l exch def newpath l b moveto l t lineto r t lineto r b lineto closepath patternNone not { ifill } if brushNone not { istroke } if end } dup 0 4 dict put def /Text { ishow } def /idef { dup where { pop pop pop } { exch def } ifelse } def /ifill { 0 begin gsave patternGrayLevel -1 ne { fgred bgred fgred sub patternGrayLevel mul add fggreen bggreen fggreen sub patternGrayLevel mul add fgblue bgblue fgblue sub patternGrayLevel mul add setrgbcolor eofill } { eoclip originalCTM setmatrix pathbbox /t exch def /r exch def /b exch def /l exch def /w r l sub ceiling cvi def /h t b sub ceiling cvi def /imageByteWidth w 8 div ceiling cvi def /imageHeight h def bgred bggreen bgblue setrgbcolor eofill fgred fggreen fgblue setrgbcolor w 0 gt h 0 gt and { l b translate w h scale w h true [w 0 0 h neg 0 h] { patternproc } imagemask } if } ifelse grestore end } dup 0 8 dict put def /istroke { gsave brushDashOffset -1 eq { [] 0 setdash 1 setgray } { brushDashArray brushDashOffset setdash fgred fggreen fgblue setrgbcolor } ifelse brushWidth setlinewidth originalCTM setmatrix stroke grestore } def /ishow { 0 begin gsave fgred fggreen fgblue setrgbcolor /fontDict printFont printSize scalefont dup setfont def /descender fontDict begin 0 [FontBBox] 1 get FontMatrix end transform exch pop def /vertoffset 1 printSize sub descender sub def { 0 vertoffset moveto show /vertoffset vertoffset printSize sub def } forall grestore end } dup 0 3 dict put def /patternproc { 0 begin /patternByteLength patternString length def /patternHeight patternByteLength 8 mul sqrt cvi def /patternWidth patternHeight def /patternByteWidth patternWidth 8 idiv def /imageByteMaxLength imageByteWidth imageHeight mul stringLimit patternByteWidth sub min def /imageMaxHeight imageByteMaxLength imageByteWidth idiv patternHeight idiv patternHeight mul patternHeight max def /imageHeight imageHeight imageMaxHeight sub store /imageString imageByteWidth imageMaxHeight mul patternByteWidth add string def 0 1 imageMaxHeight 1 sub { /y exch def /patternRow y patternByteWidth mul patternByteLength mod def /patternRowString patternString patternRow patternByteWidth getinterval def /imageRow y imageByteWidth mul def 0 patternByteWidth imageByteWidth 1 sub { /x exch def imageString imageRow x add patternRowString putinterval } for } for imageString end } dup 0 12 dict put def /min { dup 3 2 roll dup 4 3 roll lt { exch } if pop } def /max { dup 3 2 roll dup 4 3 roll gt { exch } if pop } def /midpoint { 0 begin /y1 exch def /x1 exch def /y0 exch def /x0 exch def x0 x1 add 2 div y0 y1 add 2 div end } dup 0 4 dict put def /thirdpoint { 0 begin /y1 exch def /x1 exch def /y0 exch def /x0 exch def x0 2 mul x1 add 3 div y0 2 mul y1 add 3 div end } dup 0 4 dict put def /subspline { 0 begin /movetoNeeded exch def y exch get /y3 exch def x exch get /x3 exch def y exch get /y2 exch def x exch get /x2 exch def y exch get /y1 exch def x exch get /x1 exch def y exch get /y0 exch def x exch get /x0 exch def x1 y1 x2 y2 thirdpoint /p1y exch def /p1x exch def x2 y2 x1 y1 thirdpoint /p2y exch def /p2x exch def x1 y1 x0 y0 thirdpoint p1x p1y midpoint /p0y exch def /p0x exch def x2 y2 x3 y3 thirdpoint p2x p2y midpoint /p3y exch def /p3x exch def movetoNeeded { p0x p0y moveto } if p1x p1y p2x p2y p3x p3y curveto end } dup 0 17 dict put def /storexyn { /n exch def /y n array def /x n array def n 1 sub -1 0 { /i exch def y i 3 2 roll put x i 3 2 roll put } for } def %%EndProlog %%BeginIdrawPrologue /arrowhead { 0 begin transform originalCTM itransform /taily exch def /tailx exch def transform originalCTM itransform /tipy exch def /tipx exch def /dy tipy taily sub def /dx tipx tailx sub def /angle dx 0 ne dy 0 ne or { dy dx atan } { 90 } ifelse def gsave originalCTM setmatrix tipx tipy translate angle rotate newpath arrowHeight neg arrowWidth 2 div moveto 0 0 lineto arrowHeight neg arrowWidth 2 div neg lineto patternNone not { originalCTM setmatrix /padtip arrowHeight 2 exp 0.25 arrowWidth 2 exp mul add sqrt brushWidth mul arrowWidth div def /padtail brushWidth 2 div def tipx tipy translate angle rotate padtip 0 translate arrowHeight padtip add padtail add arrowHeight div dup scale arrowheadpath ifill } if brushNone not { originalCTM setmatrix tipx tipy translate angle rotate arrowheadpath istroke } if grestore end } dup 0 9 dict put def /arrowheadpath { newpath arrowHeight neg arrowWidth 2 div moveto 0 0 lineto arrowHeight neg arrowWidth 2 div neg lineto } def /leftarrow { 0 begin y exch get /taily exch def x exch get /tailx exch def y exch get /tipy exch def x exch get /tipx exch def brushLeftArrow { tipx tipy tailx taily arrowhead } if end } dup 0 4 dict put def /rightarrow { 0 begin y exch get /tipy exch def x exch get /tipx exch def y exch get /taily exch def x exch get /tailx exch def brushRightArrow { tipx tipy tailx taily arrowhead } if end } dup 0 4 dict put def %%EndIdrawPrologue %I Idraw 10 Grid 2.84217e-39 0 %%Page: 1 1 Begin %I b u %I cfg u %I cbg u %I f u %I p u %I t [ 0.799705 0 0 0.799705 0 0 ] concat /originalCTM matrix currentmatrix def Begin %I Pict %I b u %I cfg u %I cbg u %I f u %I p u %I t [ 1 0 0 1 89.1002 831.2 ] concat Begin %I Elli %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.09512e-08 0.9 -0.9 1.09512e-08 584.1 0.89999 ] concat %I 79 550 18 17 Elli End Begin %I Line %I b 65535 3 0 1 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.09512e-08 0.9 -0.9 1.09512e-08 541.8 -27 ] concat %I 110 466 110 542 Line %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.09512e-08 0.9 -0.9 1.09512e-08 541.8 -27 ] concat %I 82 504 140 504 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-helvetica-bold-r-*-140-* Helvetica-Bold 14 SetF %I t [ 1.2168e-08 1 -1 1.2168e-08 35.5 66.1 ] concat %I [ (N) ] Text End End %I eop Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.23147 0.0157385 -0.0157385 -1.23147 409.218 -127.169 ] concat %I [ (Voluteer Boulevard) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 3.66661e-08 3.01333 -3.01333 3.66661e-08 57.9267 164.513 ] concat %I [ (UT Campus -- Jack Dongarra's Lab) ] Text End Begin %I Rect none SetB %I b n %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 1 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 693.5 -156 ] concat %I 17 61 177 478 Rect End Begin %I Line %I b 65535 2 1 1 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 6.9812e-09 0.698798 -0.573736 8.50294e-09 453.281 -79.7194 ] concat %I 158 545 1015 545 Line %I 1 End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 333.274 -81.9323 ] concat %I 257 403 4 4 Elli End Begin %I Line %I b 65535 2 0 1 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 7.02094e-09 0.702776 -0.577002 8.55135e-09 455.344 -80.5588 ] concat %I 211 544 211 17 Line %I 1 End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 6.9812e-09 0.698798 -0.573736 8.50294e-09 455.594 -124.448 ] concat %I 5 503 545 514 535 546 529 628 529 628 529 5 BSpl %I 1 End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 333.274 76.6169 ] concat %I 257 403 4 4 Elli End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 403.382 -81.1611 ] concat %I 257 403 4 4 Elli End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.15378e-08 1.1549 -0.948215 1.40528e-08 222.192 481.065 ] concat %I [ (DOWN) (TOWN) ] Text End Begin %I Poly %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.75 SetP %I t [ 2.43189e-09 0.222447 -0.19986 2.70672e-09 310.54 254.841 ] concat %I 4 301 50 850 50 850 537 301 537 4 Poly End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 9 251 541 251 516 251 507 257 495 263 489 413 491 472 496 485 499 485 499 9 BSpl %I 1 End Begin %I BSpl %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.18822e-09 0.43379 -0.755115 5.27834e-09 620.126 111.397 ] concat %I 6 486 498 502 500 510 504 514 514 514 529 513 542 6 BSpl %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2378 ] concat %I 476 654 476 542 Line %I 1 End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 445.797 -81.1611 ] concat %I 257 403 4 4 Elli End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 252.947 159.612 ] concat %I [ (Volunteer Boulevard) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 310.947 152.672 ] concat %I [ (Neyland Drive) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 198.965 155.756 ] concat %I [ (Cumberland Avenue) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 130.329 90.977 ] concat %I [ (Interstate 40) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 130.329 403.305 ] concat %I [ (Interstate 40) ] Text End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 332.433 366.124 ] concat %I 257 403 4 4 Elli End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 107 542 486 541 Line %I 1 End Begin %I Line %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 487 541 664 541 Line %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 665 541 682 540 Line %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -51.4666 ] concat %I 681 599 681 485 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.931845 2.26773e-08 -2.26773e-08 -0.931845 205.214 462.905 ] concat %I [ (Henley Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.931845 2.26773e-08 -2.26773e-08 -0.931845 193.647 318.042 ] concat %I [ (17th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 110.137 205.729 ] concat %I [ (17th Street Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 110.137 26.0435 ] concat %I [ (Airport/Alcoa Highway Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-times-bold-r-*-140-* Times-Bold 14 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 171.727 80.0262 ] concat %I [ (Cumberland Avenue Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 265.915 92.3651 ] concat %I [ (Neyland Drive Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 110.137 509.574 ] concat %I [ (Summit Hill Exit) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 144 661 155 636 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 372 662 361 634 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 752 660 736 634 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2378 ] concat %I 186 466 160 486 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2378 ] concat %I 180 576 157 542 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -5.99997 ] concat %I 475 50 328 492 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -5.99994 ] concat %I 475 483 329 589 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -6 ] concat %I 962 483 450 588 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -6 ] concat %I 450 491 474 471 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 456.136 36.7289 ] concat %I [ (To Airport) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 137.137 639.729 ] concat %I [ (To Ashville, Bristol) ] Text End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ -0.288762 0.966405 -0.966405 -0.288762 1070.92 66.5381 ] concat %I 743 193 51 70 Elli End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 600.094 464.292 ] concat %I 11 203 116 212 116 212 140 226 140 226 122 222 122 222 102 212 102 212 110 203 110 203 113 11 Poly End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.0258e-08 0.843029 -0.843029 1.0258e-08 619.253 516.221 ] concat %I 12 183 114 183 124 193 124 193 135 205 135 205 125 248 90 242 83 239 86 231 76 194 106 194 114 12 Poly End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.11572e-08 0.810855 -0.916932 9.86645e-09 659.418 526.458 ] concat %I 8 251 106 251 126 300 126 300 126 300 116 289 116 289 106 289 106 8 Poly End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.63639e-08 1.34483 -1.34483 1.63639e-08 1018.44 -282.452 ] concat %I 791 383 799 395 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 900.11 -9.08362 ] concat %I 12 776 339 776 352 798 352 803 356 808 350 807 348 812 342 808 336 803 341 789 343 782 333 776 333 12 Poly End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.13408e-08 0.952586 -0.932017 1.1591e-08 882.447 11.9457 ] concat %I 885 359 901 436 Rect End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 1 SetP %I t [ -0.249802 0.836016 -0.836016 -0.249802 1016.8 155.898 ] concat %I 743 193 51 70 Elli End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.75 SetP %I t [ -0.289327 0.966236 -0.966236 -0.289327 1071.6 80.2259 ] concat %I 707 160 754 232 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 0.240969 0.462035 -0.606024 0.318423 615.648 549.235 ] concat %I 13 164 162 182 162 182 167 235 167 234 162 254 162 254 134 234 133 235 129 183 129 183 134 164 134 164 149 13 Poly End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 9.95725e-09 0.885621 -0.818318 1.07762e-08 598.215 291.204 ] concat %I 385 148 422 197 Rect End Begin %I Pict %I b u %I cfg Black 0 0 0 SetCFg %I cbg u %I f u %I p u %I t [ 1.05665 0 0 1.05665 213.224 -6.32959 ] concat Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg DkGray 0.501961 0.501961 0.501961 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.1286e-09 0.496511 -0.668033 6.04153e-09 312.332 583.056 ] concat %I 13 204 92 204 123 226 123 226 113 248 113 248 123 264 123 264 92 248 92 248 95 226 95 226 92 225 92 13 Poly End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg DkGray 0.501961 0.501961 0.501961 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.1286e-09 -0.496511 -0.668033 -6.04153e-09 312.332 845.214 ] concat %I 13 204 92 204 123 226 123 226 113 248 113 248 123 264 123 264 92 248 92 248 95 226 95 226 92 225 92 13 Poly End Begin %I Rect none SetB %I b n %I cfg DkGray 0.501961 0.501961 0.501961 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.1286e-09 0.496511 -0.668033 6.04153e-09 311.664 582.559 ] concat %I 258 92 274 121 Rect End End %I eop Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 532.584 731.01 ] concat %I [ (Physics) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 555.924 783.814 ] concat %I [ (Geography) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 565.058 787.873 ] concat %I [ (& Geology) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 0.737179 0.804195 -0.804195 0.737179 504.155 697.286 ] concat %I [ (Biology) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 366.956 772.951 ] concat %I [ (13th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 511.152 538.704 ] concat %I [ (Voluteer Boulevard) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 0.46602 0.986397 -0.986397 0.466021 521.533 631.373 ] concat %I [ (Middle Way) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 373.044 540.56 ] concat %I [ (16th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 363.799 672.151 ] concat %I [ (Walters) (Life) (Sciences) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 537.034 899.133 ] concat %I [ (Daughtery) (Engineering) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.60693e-08 1.32061 -1.32061 1.60693e-08 657.627 707.825 ] concat %I [ (Neyland) (Stadium) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.0536 -0.283024 0.283024 -1.0536 624.893 639.464 ] concat %I [ (Stadium Drive) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 450.384 486.441 ] concat %I [ (Library) ] Text End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.645 -1.75617 ] concat %I 4 483 431 523 431 523 391 481 389 4 Poly End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 412.866 555.498 ] concat %I [ (University) ( Center) ] Text End Begin %I BSpl %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.438 0.0836792 ] concat %I 6 753 467 837 468 841 464 846 464 843 420 841 419 6 BSpl %I 1 End Begin %I BSpl %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.921 0.182861 ] concat %I 6 788 313 830 316 840 348 841 386 843 417 843 418 6 BSpl %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.435 0.47699 ] concat %I 887 450 839 438 Line %I 1 End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.589 0.222778 ] concat %I 9 807 460 838 460 838 405 816 404 816 394 831 394 831 379 807 378 806 379 9 Poly End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.838 0.0155029 ] concat %I 796 369 781 389 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 523.481 806.106 ] concat %I [ (South) (College) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 1.21714e-08 1.00029 -1.00029 1.21714e-08 439.801 711.069 ] concat %I [ (Ayres Hall) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 771.842 61.8701 ] concat %I 943 284 915 303 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 459.421 900.429 ] concat %I [ (Dabney/) (Buhler) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 767.413 57.441 ] concat %I 698 424 663 374 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-times-bold-r-*-140-* Times-Bold 14 SetF %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 464.596 743.589 ] concat %I [ (X) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09087 0.0128026 -0.0128027 -1.09087 503.693 612.151 ] concat %I [ (Stadium Drive) ] Text End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < cc cc 33 33 cc cc 33 33 > -1 SetP %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 855.109 167.282 ] concat %I 465 390 498 427 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 6.73658e-10 0.0553633 -0.0553633 6.73658e-10 494.25 582.008 ] concat %I 11 509 1018 237 1018 237 1114 -35 1114 -35 1018 -307 1018 -307 746 -35 746 -35 378 509 378 509 380 11 Poly End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 485.903 693.513 ] concat %I [ (Psychology) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 9.87788e-09 0.811792 -0.811792 9.87788e-09 526.522 575.313 ] concat %I [ (Parking) (Garage) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < cc cc 33 33 cc cc 33 33 > -1 SetP %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 774.943 159.309 ] concat %I 481 283 491 300 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 368.693 657.005 ] concat %I [ (15th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 374.042 893.422 ] concat %I [ (11th Street) ] Text End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 18 476 304 494 301 501 299 527 283 545 269 569 255 590 244 611 241 648 238 672 234 686 219 705 203 741 204 767 217 776 236 774 284 776 343 773 538 18 BSpl %I 1 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 376 429 375 538 Line %I 1 End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 8 376 429 375 275 369 238 352 223 345 219 323 208 303 203 303 204 8 BSpl %I 1 End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 4 301 50 850 50 850 537 301 537 4 Poly End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -5.99995 ] concat %I 5 814 52 872 66 910 82 947 107 961 127 5 BSpl %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.520591 0.958718 -0.958718 -0.520591 712.173 866.792 ] concat %I [ (Neyland Drive) ] Text End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 653.5 ] concat %I 24 459 412 422 314 224 273 224 233 224 152 305 80 345 72 386 31 418 -33 418 -122 394 -316 474 -461 652 -542 854 -566 1015 -566 1152 -501 1184 -445 1176 -203 1168 15 1031 233 789 314 547 314 432 351 428 350 24 BSpl %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 653.5 ] concat %I 916 415 915 1204 Line %I 8 End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 620.75 ] concat %I 5 486 195 402 153 394 72 402 -73 394 -73 5 BSpl %I 8 End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 604.375 ] concat %I 11 133 415 132 396 110 602 231 715 316 751 330 765 387 857 351 977 344 1126 358 1041 351 1190 11 BSpl %I 8 End Begin %I MLine %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 604.375 ] concat %I 3 133 410 153 -454 613 -2226 3 MLine %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 604.375 ] concat %I 663 276 137 276 Line %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 527 522.5 ] concat %I 96 105 808 79 Line %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 527 522.5 ] concat %I 95 105 -425 89 Line %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 483.5 457 ] concat %I 99 587 3984 588 Line %I 8 End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 6.91136e-09 0.653301 -0.567997 7.94935e-09 450.164 -70.1652 ] concat %I 15 211 347 224 320 247 286 278 265 315 254 368 252 499 251 582 255 629 257 783 266 863 302 880 318 903 384 900 434 898 545 15 BSpl %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 1.2168e-08 1 -1 1.2168e-08 288.5 532.5 ] concat %I [ (Jack Dongarra's office in Ayres Hall Room 107) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ -0.792068 1.92757e-08 -1.92757e-08 -0.792068 428.69 73.2605 ] concat %I [ (Airport/Alcoa Highway) ] Text End Begin %I Rect %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < 88 44 22 11 88 44 22 11 > -1 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 610 221 ] concat %I 258 413 267 424 Rect End Begin %I Rect %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < 88 44 22 11 88 44 22 11 > -1 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 593 221 ] concat %I 258 413 267 424 Rect End Begin %I MLine %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 3 558 602 521 602 507 598 3 MLine %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 326 526 326 467 Line %I 1 End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 564 368 588 385 Rect End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 773 26.9999 ] concat %I 564 368 588 385 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.96587e-09 0.736842 -0.736842 8.96587e-09 708.816 120.079 ] concat %I 4 616 414 631 414 631 431 616 431 4 Poly End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.855758 0.676646 -0.676646 -0.855758 381.905 599.524 ] concat %I [ (Law Builfinh) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.855758 0.676646 -0.676646 -0.855758 383.905 547.524 ] concat %I [ (Pan-Helenic Bldg.) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.855758 0.676646 -0.676646 -0.855758 384.905 572.524 ] concat %I [ (International House.) ] Text End Begin %I MLine %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 2 559 582 507 581 2 MLine %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 186.965 539.756 ] concat %I [ (Ramada Inn) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 167.965 538.756 ] concat %I [ (Hilton) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 1.2168e-08 1 -1 1.2168e-08 514.5 56.5 ] concat %I [ (Directions from the airport to Ayres Hall:) () ( Alcoa Highway North to Cumberland Avenue) () ( Cumberland Avenue east to Stadium Drive) ( \(Stadium Dr. is accross from 15th St.\)) () ( Park at Parking Garage and walk up hill) ( to largest building, Ayres Hall) ] Text End End %I eop showpage %%Trailer end From owner-pbwg-comm@CS.UTK.EDU Mon Feb 22 10:36:00 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA05312; Mon, 22 Feb 93 10:36:00 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA28038; Mon, 22 Feb 93 10:34:47 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 22 Feb 1993 10:34:46 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from antares.mcs.anl.gov by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA28025; Mon, 22 Feb 93 10:34:42 -0500 Received: from jadoube.mcs.anl.gov by antares.mcs.anl.gov with SMTP id AA23129 (5.65c/IDA-1.4.4 for ); Mon, 22 Feb 1993 09:34:38 -0600 From: David Levine Received: by jadoube.mcs.anl.gov (4.1/GeneV4) id AA11892; Mon, 22 Feb 93 09:34:37 CST Date: Mon, 22 Feb 93 09:34:37 CST Message-Id: <9302221534.AA11892@jadoube.mcs.anl.gov> To: pbwg-comm@cs.utk.edu send archive from pbwg From owner-pbwg-comm@CS.UTK.EDU Mon Feb 22 15:29:10 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA15809; Mon, 22 Feb 93 15:29:10 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA15065; Mon, 22 Feb 93 15:27:47 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 22 Feb 1993 15:27:45 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from adios.brl.mil by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA15057; Mon, 22 Feb 93 15:27:44 -0500 Date: Mon, 22 Feb 93 15:24:05 EST From: TomC To: pbwg-comm@CS.UTK.EDU Cc: TomC , coleman@BRL.MIL, apress@BRL.MIL Subject: Additions to pbwg maillist Message-Id: <9302221524.aa18550@ADIOS.BRL.MIL> Please include the following in the above mail list crimmins@brl.mil Tom Crimmins 410-278-6267 monte@brl.mil Monte Coleman 410-278-6261 apress@brl.mil Tony Pressley 410-278-6509 Thanks, Tom Crimmins From owner-pbwg-comm@CS.UTK.EDU Mon Feb 22 17:04:11 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA19520; Mon, 22 Feb 93 17:04:11 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA19497; Mon, 22 Feb 93 17:03:39 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 22 Feb 1993 17:03:38 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from climate.engin.umich.edu by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA19489; Mon, 22 Feb 93 17:03:36 -0500 Received: by climate.engin.umich.edu (5.67/1.35) id AA21283; Mon, 22 Feb 93 17:05:11 -0500 Date: Mon, 22 Feb 93 17:05:11 -0500 From: Hal G Marshall Message-Id: <9302222205.AA21283@climate.engin.umich.edu> To: pbwg-comm@cs.utk.edu Cc: idaho@CS.UTK.EDU Please add me to your list for MIMD benchmarks. -Hal From owner-pbwg-comm@CS.UTK.EDU Mon Mar 1 05:14:41 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA28056; Mon, 1 Mar 93 05:14:41 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA15881; Mon, 1 Mar 93 05:13:28 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 1 Mar 1993 05:13:27 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from Mail.Think.COM by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA15867; Mon, 1 Mar 93 05:13:24 -0500 Received: from Godot.Think.COM by mail.think.com; Mon, 1 Mar 93 05:13:22 -0500 Received: by godot.think.com (4.1/Think-1.2) id AA16751; Mon, 1 Mar 93 05:14:28 EST Message-Id: <9303011014.AA16751@godot.think.com> To: Jack Dongarra Cc: pbwg-comm@cs.utk.edu Subject: Re: Parallel Benchmark Working Group Meeting In-Reply-To: Your message of "Sun, 14 Feb 93 16:56:01 EST." <9302142156.AA01941@dasher.cs.utk.edu> Date: Mon, 01 Mar 93 05:14:28 EST From: Dennis Parkinson I apologize I will not be able to attend the meeting sorry for the late notifia`cation dennis Parkinson From owner-pbwg-comm@CS.UTK.EDU Mon Mar 1 07:39:16 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA25550; Mon, 1 Mar 93 07:39:16 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA25188; Mon, 1 Mar 93 07:38:13 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 1 Mar 1993 07:38:12 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from inetg1.Arco.COM by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA25180; Mon, 1 Mar 93 07:38:11 -0500 Received: by Arco.COM (4.1/SMI-4.1) id AA15125; Mon, 1 Mar 93 06:38:08 CST Date: Mon, 1 Mar 93 06:38:08 CST From: ccm@Arco.COM (Chuck Mosher (214)754-6468) Message-Id: <9303011238.AA15125@Arco.COM> To: pbwg-comm@cs.utk.edu Subject: Addition to the List Please add me to the parallel benchmark working group mailing list. Regards, Chuck Mosher ccm@arco.com From owner-pbwg-comm@CS.UTK.EDU Mon Mar 1 10:49:36 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA29573; Mon, 1 Mar 93 10:49:36 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA05800; Mon, 1 Mar 93 10:48:55 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 1 Mar 1993 10:48:54 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from ben.uknet.ac.uk by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA05792; Mon, 1 Mar 93 10:48:45 -0500 Received: from eros.uknet.ac.uk by ben.uknet.ac.uk via UKIP with SMTP (PP) id ; Mon, 1 Mar 1993 15:48:23 +0000 Received: from parsys.co.uk by eros.uknet.ac.uk with UUCP id <9375-0@eros.uknet.ac.uk>; Mon, 1 Mar 1993 15:48:01 +0000 Message-Id: <2594.9303011242@boston.parsys.co.uk> Received: from parsys.co.uk (gmail) by boston.parsys.co.uk; Mon, 1 Mar 93 12:42:29 GMT Date: 1 Mar 1993 12:23:18 +0000 From: Heather Cooke Subject: Re: Parallel Benchmark Worki To: dongarra@cs.utk.edu, System_Manager@parsys.co.uk Cc: pbwg-comm@cs.utk.edu Mail*Link(r) SMTP RE>Parallel Benchmark Worki Received: by parsys.co.uk with SMTP;1 Mar 1993 11:49:21 +0000 Received: from uknet.UUCP by boston.parsys.co.uk; Mon, 1 Mar 93 11:45:35 GMT Received: from CS.UTK.EDU by ben.uknet.ac.uk via EUnet with SMTP (PP) id ; Mon, 1 Mar 1993 10:17:20 +0000 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA15881; Mon, 1 Mar 93 05:13:28 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 1 Mar 1993 05:13:27 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from Mail.Think.COM by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA15867; Mon, 1 Mar 93 05:13:24 -0500 Received: from Godot.Think.COM by mail.think.com; Mon, 1 Mar 93 05:13:22 -0500 Received: by godot.think.com (4.1/Think-1.2) id AA16751; Mon, 1 Mar 93 05:14:28 EST Message-Id: <9303011014.AA16751@godot.think.com> To: Jack Dongarra Cc: pbwg-comm@cs.utk.edu Subject: Re: Parallel Benchmark Working Group Meeting In-Reply-To: Your message of "Sun, 14 Feb 93 16:56:01 EST." <9302142156.AA01941@dasher.cs.utk.edu> Date: Mon, 01 Mar 93 05:14:28 EST From: Dennis Parkinson I apologize I will not be able to attend the meeting sorry for the late notifia`cation dennis Parkinson From owner-pbwg-comm@CS.UTK.EDU Wed Mar 3 13:13:23 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA05199; Wed, 3 Mar 93 13:13:23 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA25195; Wed, 3 Mar 93 13:12:14 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 3 Mar 1993 13:12:12 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from orion.oac.uci.edu by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA25186; Wed, 3 Mar 93 13:12:09 -0500 Received: from balboa.eng.uci.edu by orion.oac.uci.edu with SMTP id AA14344 (5.65c/IDA-1.4.4 for ); Wed, 3 Mar 1993 10:12:00 -0800 Received: from seal.eng.uci.edu by balboa.eng.uci.edu (4.1/SMI-4.1) id AA03526; Wed, 3 Mar 93 10:12:02 PST From: mnyeu@balboa.eng.uci.edu (Maung Nyeu) Message-Id: <9303031812.AA03526@balboa.eng.uci.edu> Subject: Parallel Benhmarking Meetinh To: pbwg-comm@cs.utk.edu Date: Wed, 3 Mar 93 10:10:46 PST X-Mailer: ELM [version 2.3 PL11] Hi, I am student at UC Irvine dept. of comp science. I am interested to get the summary or the papers presented in the parallel benchmarking meeting. I would appreciate it if you send me a reply. Thanks. -- Maung Ting Nyeu email: mnyeu@balboa.eng.uci.edu From owner-pbwg-comm@CS.UTK.EDU Thu Mar 4 12:21:42 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA02456; Thu, 4 Mar 93 12:21:42 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA06000; Thu, 4 Mar 93 12:20:35 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Thu, 4 Mar 1993 12:20:33 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from timbuk.cray.com by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA05984; Thu, 4 Mar 93 12:20:31 -0500 Received: from groucho (groucho.cray.com) by cray.com (4.1/CRI-MX 2.13) id AA06976; Thu, 4 Mar 93 11:20:29 CST Received: by groucho (4.1/CRI-5.13) id AA16192; Thu, 4 Mar 93 11:20:26 CST From: kkg@ferrari.cray.com (Koushik Ghosh) Message-Id: <9303041720.AA16192@groucho> Subject: Meeting on May 24, 1993 To: pbwg-comm@cs.utk.edu Date: Thu, 4 Mar 93 11:20:25 CST X-Mailer: ELM [version 2.3 PL11] There seems to be a lot of requests for a May 24, 1993 meeting. May I suggest that if we start the meeting at 8 AM instead of 9 AM, the chances of getting everything done by 5PM might go up. Then Jack and some of us might be able to catch a flight out of Knoxville that evening. -- Koushik Ghosh kkg@cray.com Benchmarking Dept. Cray Research Inc. 1-612-683-3407 From owner-pbwg-comm@CS.UTK.EDU Thu Mar 4 13:48:21 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA05140; Thu, 4 Mar 93 13:48:21 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA11235; Thu, 4 Mar 93 13:47:13 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Thu, 4 Mar 1993 13:47:10 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from gemini.npac.syr.EDU by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA11220; Thu, 4 Mar 93 13:47:05 -0500 Received: by gemini.npac.syr.edu id AA20470 (5.65c/IDA-1.4.4 for pbwg-comm@cs.utk.edu); Thu, 4 Mar 1993 13:46:43 -0500 Date: Thu, 4 Mar 1993 13:46:43 -0500 From: Tomasz Haupt Message-Id: <199303041846.AA20470@gemini.npac.syr.edu> To: pbwg-comm@cs.utk.edu Subject: Meeting on May 24, 1993 Cc: haupt@npac.syr.edu > There seems to be a lot of requests for a May 24, 1993 meeting. Fine with me. Tom Haupt From owner-pbwg-comm@CS.UTK.EDU Fri Mar 5 13:15:58 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA04015; Fri, 5 Mar 93 13:15:58 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA28005; Fri, 5 Mar 93 13:14:48 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Fri, 5 Mar 1993 13:14:46 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from vnet.ibm.com by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA27996; Fri, 5 Mar 93 13:14:32 -0500 Message-Id: <9303051814.AA27996@CS.UTK.EDU> Received: from KGNVMZ by vnet.ibm.com (IBM VM SMTP V2R2) with BSMTP id 6262; Fri, 05 Mar 93 13:11:37 EST Date: Fri, 5 Mar 93 12:45:31 EST From: "Dr. Joanne L. Martin" To: pbwg-comm@cs.utk.edu Subject: send archive from pbwg send archive from pbwg From owner-pbwg-comm@CS.UTK.EDU Tue Mar 9 15:08:55 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA13515; Tue, 9 Mar 93 15:08:55 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA24030; Tue, 9 Mar 93 15:01:29 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 9 Mar 1993 15:01:28 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from THUD.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA24023; Tue, 9 Mar 93 15:01:27 -0500 From: Jack Dongarra Received: by thud.cs.utk.edu (5.61++/2.7c-UTK) id AA14193; Tue, 9 Mar 93 15:01:25 -0500 Date: Tue, 9 Mar 93 15:01:25 -0500 Message-Id: <9303092001.AA14193@thud.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: PBWG posting Forwarding: Mail from '' dated: Tue, 9 Mar 93 14:30:05 -0500 Here is a note I'm planning on posting to comp.parallel. Let me know if you see any problems. I will post it on Friday. 3/12. Regards, Jack Dear Colleague, We are planning to have the Third Meeting of the Parallel Benchmark Working Group meet in Knoxville, Tennessee at the University of Tennessee on May 24th, 1993. This process formally began with a workshop held at the Supercomputer '92 meeting in November 1992. The purpose of the working group is to establish credible and useful benchmarks for the evaluation of Distributed Memory MIMD systems. The objectives for the group are: 1. To establish a comprehensive set of parallel benchmarks that is generally accepted by both users and vendors of parallel system. 2. To provide a focus for parallel benchmark activities and avoid unnecessary duplication of effort and proliferation of benchmarks. 3. To set standards for benchmarking methodology and result-reporting together with a control database/repository for both the benchmarks and the results. Mode of Working: The working group has adopted an HPF-like forum style of proceedings with a view to convergence to an agreed set of benchmarks and procedures within 10 months. If you would like to participate and attend the meeting let me know. Mailing Lists ============= The following mailing lists have been set up. pbwg-comm@cs.utk.edu Whole committee pbwg-lowlevel@cs.utk.edu Low level subcommittee pbwg-compactapp@cs.utk.edu Compact applications subcommittee pbwg-method@cs.utk.edu Methodology subcommittee pbwg-kernel@cs.utk.edu Kernel subcommittee If you are on a mailing list you will receive mail as it is posted. If you want to join a mailing list send me mail (dongarra@cs.utk.edu). All mail will be collected and can be retrieved by sending email to netlib@ornl.gov and in the mail message typing: send comm.archive from pbwg send lowlevel.archive from pbwg send compactapp.archive from pbwg send method.archive from pbwg send kernel.archive from pbwg send index from pbwg The various subcommittees will look into the following topics: Low-Level: --------- Start-up, latency, bandwidth Reduction (broadcast, sum, gather/scatter) Synchronization (e.g., SYNCH1 from Genesis) I/O Kernel: ------ Matrix operations (e.g., multiply, transpose) LU Decomposition PDE Solvers (Red/Black Relaxation) Multigrid FFT Conjugate Gradient Compact Applications: -------------------- Particle-In-Cell codes (e.g., LPM1 from Genesis) QCD Molecular Dynamics CFD ARCO Financial Applications Methodology: ------------ Guidelines for reporting performance. The meeting site will be the: University Center in room 221 University of Tennessee We have made arrangements with the Hilton Hotel in Knoxville. Hilton Hotel 501 W. Church Street Knoxville, TN Phone: 615-523-2300 When making arrangements tell the hotel you are associated with the Parallel Benchmarking Meeting. You can rent a car or get a cab from the airport to the hotel. >From the hotel to the University it is a 15 minute walk. We should plan to start at 8:30 pm May 24th and finish about 5:00 pm. The format of the meeting is: Monday 25th May 8.30 - 12.00 Full group meeting 12.00 - 1.30 Formal lunch 1.30 - 4.00 Parallel subgroup meetings 5.00 - 5.00 Full group meeting Tentative agenda for full group meeting: 1. Minutes of Minneapolis meeting 2. Reports and discussion from subgroups 3. Open discussion and agreement on further actions 4. Date and venue for next meeting Suggested subgroups - probably two in parallel Compact Applications Low-Level benchmarks and second pair: Kernels benchmarks Methodology We have setup a mail refector for correspondence, it is called pbwg-comm@cs.utk.edu. Mail to that address will be sent to the mailing list and also collected in netlib@ornl.gov. To retrieve the collected mail, send email to netlib@ornl.gov and in the mail message type: send comm.archive from pbwg If you would like to be put on the mailing list to receive the correspondence let me know. Regards, Jack Dongarra ----------- End Forwarded Message ----------- From owner-pbwg-comm@CS.UTK.EDU Tue Mar 9 17:22:57 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA15724; Tue, 9 Mar 93 17:22:57 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA01135; Tue, 9 Mar 93 17:21:35 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 9 Mar 1993 17:21:32 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from BERRY.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61++/2.8s-UTK) id AA01122; Tue, 9 Mar 93 17:21:29 -0500 From: Mike Berry Received: by berry.cs.utk.edu (5.61++/2.7c-UTK) id AA24206; Tue, 9 Mar 93 17:19:08 -0500 Date: Tue, 9 Mar 93 17:19:08 -0500 Message-Id: <9303092219.AA24206@berry.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: PBWG Minutes (March 1-2, 1993) Minutes of the 2nd Meeting of the Parallel Benchmark Working Group (PBWG) ------------------------------------------------------------------------- Place: Room 221, University Center University of Tennessee Knoxville, TN Host: Jack Dongarra ORNL/Univ. of Tennessee Dates: March 1-2, 1993 Attendees/Affiliations: ---------------------- David Bailey, NASA Michael Berry, Univ. of Tennessee Ed Brocklehurst, National Physical Lab Jack Dongarra, Univ. of Tennessee / ORNL Koushik Ghosh, Cray Research Tom Haupt, Syracuse Univ. Tony Hey, Southampton Univ. Roger Hockney, Southampton Univ. Brian LaRose, Univ. of Tennessee David Mackay, Intel SSD Joanne Martin, IBM Robert Pennington, Pittsburgh Supercomputing Center David Walker, ORNL Pearl Wang, George Mason Univ. / US Geological Survey Agenda: March 1, 1993 --------------------- At 1:00 pm EDT, Roger Hockney gave opening remarks and welcomed all participants to the workshop. Each participant introduced him or herself by affiliation and interests. The minutes of the previous meeting at Supercomputing'92 (Minneapolis) were reviewed. Roger suggested that the group consider an alternative name for the group such as "PARABEN" or "INTERBEN". Participants were asked to think about this over dinner and cocktails that evening. Jack Dongarra mentioned that the mail reflector "pbwg-com@cs.utk.edu" is now in place and can be used as an electronic forum for the group. Currently there are about 50 persons on the mailing list. Jack also mentioned that anyone can obtain all correspondence information concerning PBWG requests and solicitations by sending the mail message "send archive from pbwg" to netlib@ornl.gov. To inquire about the contents of the archive simply send the message "send index from pbwg" to netlib@ornl.gov. It was noted that this meeting was posted in the the "comp.parallel" users group (USENET). Roger Hockney reviewed the 4 subgroups previously defined for the effort: Methodology, Classification, Compiler-based, Kernels. Roger proposed that the participants not break into subgroups due to the somewhat lower than expected turnout for the workshop. All participants agreed and a general discussion on "Methodology" followed. David Bailey was then asked to review the "Proposed Guidelines for Reporting Performance" which he distributed to the group. [These guidelines will be made available from the pbwg archive in netlib.] Roger Hockney then distributed an article entitled "A framework for benchmark performance analysis" from Supercomputer 48, IX-2 (March 1992). He recommended that the group start using correct performance symbols such as "Mflop/s" for millions of floating-point operations per second rather than simply "Mflops" which is somewhat ambiguous. As discussed in the paper, temporal performance measures such as T(N,P) and R_T=T^-1(N,P) were also suggested. Roger pointed out that the numerator term, F(N), in the performance measure R_B = F(N)/T(N,P) should be constant and clearly stated. Here, T(N,P) is the execution time, R_T is the inverse of execution time, N is the problem size, P is the number of processors, and F(N) is the total flop count. He suggested that the group steer away from measures such as speedup and efficiency. David Bailey pointed out that one must be prepared to change F(N) to match optimal implementations on different machines. Jack D. pointed out that one must be careful though not to allow performance rates which exceed peak performance. Roger H. suggested that the performance database might provide plots of the metrics as functions of N and P. Along with the distribution of the paper by Roger Hockney, Tony Hey distributed copies of 2 papers on the Genesis Benchmarks: "The Genesis Distributed-Memory Benchmarks" (Tech Report, Dept. of Electronics and Computer Science, Univ. of Southhampton, UK), and "The Genesis distributed-memory benchmarks. Part 1: methodology and general relativity benchmark with results for the SUPRENUM computer" (from Concurrency: Practice and Experience Vol. 5(1), pp. 1-22, February 1993.) Tony indicated that the Genesis benchmarks will be sent to Jack D. for inclusion in the benchmark repository in NETLIB. Ed Brocklehurst raised the concern that the European PEPS project methodology might differ from that of PBWG. He stressed the need for collaboration among the two efforts. He then gave a short lecture on the PEPS project which is a 3-year funded program ($11 million) which began in November of 1992. Participants of the PEPS project include: TS-ASM (electronics firm - France), NPL, Inecs Fistemi (Italy), Simulog (France), Sosip (Germany, France, Italy, UK), and Warwick. The benchmarks supported by PEPS are to be available to the public. Other concerns of the project involve performance monitoring and characterization. The discussion on Methodology concluded with a comment by Michael Berry that all accepted codes for the benchmark suite be self-validating. Roger H. then asked David Walker to report on the work thus far by the Classification subgroup. David W. stressed that the classification of benchmarks be based on (i.) communication characteristics, (ii.) I/O, and (iii.) load balance and data layout. Questions on the inclusion of profiles and processor utilization data with the benchmarks were raised. David W. pointed out that benchmarks may range from inherent parallel codes (e.g., Monte-Carlo) to mesh-based codes or gather/scatter-based applications. David Bailey stressed the need for classifying I/O and to see what the current measurement of I/O is in existing benchmark suites. Joanne Martin pointed out that the ARCO benchmark suite is well-received by the seismic community and should be seriously considered by PBWG. The discussion then turned to compiler-based benchmarks with Tom Haupt taking the lead. He distributed the document entitled "High Performance Fortran, Fortran-D Benchmark Suite" which is available vi anonymous ftp from minerva.npac.syr.edu. Tom H. indicated that the Syracuse compiler benchmarks can test parsers on a collection of statements from HPF, accept intrinsic functions, and for-all loops. He stressed that these benchmarks be able to handle hard-coded message-passing as well as HPF Fortran. One goal would be to find instances where HPF is difficult to use. Tom mentioned that these codes deal with parallel I/O, validate HPF Fortran, and generate meaningful statistics. The current Syracuse benchmark suite contains over 50 codes and includes many well-known benchmarks such as LAPACK, NAS, and Genesis benchmarks. Tom agreed to download this suite to machine at UT during the workshop (this was done after the meeting adjourned on March 2 and the files are available in netlib, send index from pbwg, for more details). This suite consumes about 5 megabytes and comprises about 14 directories of README files, makefiles, and source files. Brian LaRose (student of Michael B.) then gave a demonstration of the Performance Database Server (PDS) which is available in an upcoming release of XNETLIB. Beforehand, Jack D. gave a brief introduction of XNETLIB to familiarize participants with the features of the X Windows-based user interface to NETLIB. The members were pleased with the tool and made suggestions for improvements such as graphing data retrieved from the database. Brian demonstrated the various browsing, rank-ordering, and search features of the system he is designing for his Master thesis at the Univ. of Tennessee. Michael B. indicated that participants would be able to test PDS and should send anonymous ftp address/directory information to "utpds@cs.utk.edu" so a beta release of the software could be sent to them. At this point, Roger Hockney asked that Jack D. define/classify the publically-available benchmarks that might be considered for the PBWG benchmark suite. The classification into Kernels, Applications, and Compact Applications is given below. A (*) indicates availability as single-node benchmarks also. Additional benchmarks that were not explicitly classified are listed under "Miscellaneous". Kernels: ------- Linpack 100 (*) 1000 (*) Livermore Loops (*) Los Alamos (*, some of them) Tom Dunigan's (ORNL) Genesis Global Operations (e.g., sum, gather/scatter) NAS Sorting Applications: ------------ Perfect Slalom RAPS PEPS Compact Applications: -------------------- NAS Genesis Syracuse ARCO Euroben (*) Miscellaneous: ------------ flops.c IOZONE Dhrystones Whetstones Xstones Jack D. proposed that the selected codes for the PBWG suite be implemented in Fortran-77, HPF, in PVM for message-passing (this will be converted to MPI, Message Passing Interface, when the standard is complete). Roger H. stressed the need to measure startup latency (alpha, beta), and metrics such as R-sub-infinity and N-sub-one-half. Other target measures discussed included: peak rates, flop/s, communication costs, and bisection (node-node) bandwidth. David Bailey stressed that the single processor benchmarks should be separated from the parallel benchmarks too avoid confusion. At 5:00pm EDT, the workshop adjourned for a 6:30pm EDT dinner appointment at Chesapeakes's, a downtown Knoxville seafood restaurant. Agenda: March 2, 1993 --------------------- At 9:00 am EDT, Roger Hockney opened the morning session of the PBWG workshop in Room 221 of the University Center on the campus of the University of Tennessee. Prior to any technical discussion on the makeup of the PBWG benchmark suite. Participants agreed to hold a "Birds of a Feather" session at the Supercomputing'93 conference in Portland, OR in November. Joanne Martin will schedule this meeting for the group - a 4:30 pm session was agreed upon by the participants. Prior to the Supercomputing'93 meeting, Jack D. agreed to collect all desirable benchmarks and install them in NETLIB for public access. The discussion on the categorical makeup of a PBWG benchmark suite was then lead by Tony Hey. The 3 categories considered were: Low-Level, Kernels, and Compact Applications. David W. stressed that 10 would be an optimal number of codes to use. Participants generally agreed with this proposal. Potential benchmarks considered are listed by category below. Jack D. stressed that each group "justify" benchmarks by indicating exactly what machine feature(s) are to be tested and try to avoid overlap between categories. Low-Level: --------- Start-up, latency, bandwidth Reduction (broadcast, sum, gather/scatter) Synchronization (e.g., SYNCH1 from Genesis) I/O Kernel: ------ Matrix operations (e.g., multiply, transpose) LU Decomposition PDE Solvers (Red/Black Relaxation) Multigrid FFT Conjugate Gradient Compact Applications: -------------------- Particle-In-Cell codes (e.g., LPM1 from Genesis) QCD Molecular Dynamics CFD ARCO Financial Applications At this point, Roger H. asked that subgroups be formed to address the each of the above categories of benchmarks. Each subgroup will be ask to give a report at the Portland Meeting concerning the potential design/makeup of their category of PBWG benchmarks. Participants agreed that the subgroups be open to other experts from the performance community: vendors, academia, and users. A person may be in more than one subgroup and Jack D. suggested that a voting strategy be developed. A "1 vote per institution" was one candidate for consideration but nothing official was decided at the workshop. Suggestions for subgroup members not present at the workshop are included below. A (*) indicates subgroup leader. PBWG Benchmark Suite Subgroups: ------------------------------ Low-Level Kernel Compact Applications Methodology --------- ------ -------------------- ----------- PEPS/Warwick D. Walker D. Walker (*) D. Bailey (*) IBM/D. Frye J. Dongarra J. Martin D. Schneider R. Pennington T. Hey (*) E. Brocklehurst PEPS R. Hockney (*) T. Haupt K. Ghosh J. Dongarra PEPS M. Berry E. Kushner (Intel) D. Bailey D. Bailey P. Wang D. Barton The following mail reflectors have been setup: Low Level: pbwg-lowlevel@cs.utk.edu Kernel: pbwg-kernel@cs.utk.edu Compact Applications: pbwg-compactapp@cs.utk.edu Methodology: pbwg-method@cs.utk.edu Jack D. agreed to set up mail reflectors for each of the subgroups. (These mail reflectors are setup and functioning.) All participants agreed to 2 meetings that would be open to anyone interested in attending prior to the meeting at Supercomputing'93 in Portland. Drafts of the reports from each subgroup are due by the agreed-upon May 24 meeting at the Univ. of Tennessee which will begin at 8:30 am. The second meeting date will also be at the Univ. of Tennessee on August 23-24. Here's the approved schedule of PBWG meetings: Third Meeting: May 24, begins at 8:30am (Univ. of Tennessee) Fourth Meeting: August 23-24 (Univ. of Tennessee) Fifth Meeting: Week of Nov 15 (Supercomputing Portland, OR) Joanne Martin mentioned that there was a Conference on Performance Tools on April 1-4 and Tony H. suggested that PBWG might submit an announcement to that conference about PBWG's activities. A discussion on possible name changes for PBWG was considered again. Some suggestions included PES(Performance or Parallel Evaluation Suite) and PARKBENCH(Parallel Kernel Benchmarks). Since no consensus was reached, the discussion is postponed to the next meeting so members can think of other names. In preparation for the May meeting, subgroup leaders briefly mentioned some of the issues they hope to address: T. Hey : talk to vendors, provide list of kernel areas, justification D. Walker : define areas, acquire candidate codes, address scalability, address problem sizes (e.g., 4 used in Genesis), be able to specify data layouts (use vanilla layouts) R. Hockney: construct list of desirable low-level benchmarks Participants agreed that users should be able to optimize the suite by any means provided they document what they do (e.g., assembly language allowed). When possible, benchmark contributions should include the code that was used to obtain the reported results (stored in PDS). The second day of the PBWG workshop was adjourned by Roger Hockney at 11:00am EDT, and participants had informal discussions before departing the Univ. of Tennessee campus. PBWG activities will be posted to the comp.parallel newsgroup (USENET). From owner-pbwg-comm@CS.UTK.EDU Wed Mar 24 06:54:11 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA05932; Wed, 24 Mar 93 06:54:11 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA11133; Wed, 24 Mar 93 06:53:25 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 24 Mar 1993 06:53:13 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA11125; Wed, 24 Mar 93 06:53:10 -0500 Via: uk.ac.southampton.ecs; Wed, 24 Mar 1993 11:44:52 +0000 From: R.Hockney@parallel-applications-centre.southampton.ac.uk Via: calvados.pac.soton.ac.uk (plonk); Wed, 24 Mar 93 11:37:49 GMT Date: Wed, 24 Mar 93 11:24:51 GMT Message-Id: <3278.9303241124@calvados.pac.soton.ac.uk> Apparently-To: pbwg-comm@cs.utk.edu From owner-pbwg-comm@CS.UTK.EDU Wed Mar 24 07:10:28 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA09660; Wed, 24 Mar 93 07:10:28 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA11684; Wed, 24 Mar 93 07:09:56 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 24 Mar 1993 07:09:55 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA11676; Wed, 24 Mar 93 07:09:50 -0500 Via: uk.ac.southampton.ecs; Wed, 24 Mar 1993 12:07:41 +0000 From: R.Hockney@parallel-applications-centre.southampton.ac.uk Via: calvados.pac.soton.ac.uk (plonk); Wed, 24 Mar 93 11:44:01 GMT Date: Wed, 24 Mar 93 11:50:26 GMT Message-Id: <3341.9303241150@calvados.pac.soton.ac.uk> Apparently-To: pbwg-comm@cs.utk.edu ~r comlet1.asc From owner-pbwg-comm@CS.UTK.EDU Wed Mar 24 07:36:42 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA11698; Wed, 24 Mar 93 07:36:42 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA12605; Wed, 24 Mar 93 07:35:41 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 24 Mar 1993 07:35:40 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA12584; Wed, 24 Mar 93 07:35:34 -0500 Via: uk.ac.southampton.ecs; Wed, 24 Mar 1993 12:32:55 +0000 Via: brewery.ecs.soton.ac.uk; Wed, 24 Mar 93 12:25:44 GMT From: Prof Roger Hockney Received: from alithea.ecs.soton.ac.uk by brewery.ecs.soton.ac.uk; Wed, 24 Mar 93 12:34:08 GMT Date: Wed, 24 Mar 93 12:34:10 GMT Message-Id: <253.9303241234@alithea.ecs.soton.ac.uk> To: pbwg-comm@cs.utk.edu Subject: PBWG Draft Report A Message from Your Chairman ---------------------------- In view of the committee's aim to agree on a draft text at the meeting on May 24, 1993, it would help if each subcommittee produces their recommendations as LATEX files to fit into a standard framework. I believe that LATEX is the most widely used text processing system in the universities, and appears to have been used for the HPF report. I have prepared the following skeleton files for the committee's report: (1) benrep1.tex - control file for main report (2) bencom1.tex - command definition file (3) benref1.bib - start of bibliography Additions to the command file and bibliography can be sent to me as files bencom2.tex, benref2.bib etc. The control file reads the following files which are to be provided by the leader of each subcommittee: (4) intro1.tex - Roger Hockney for whole committee (5) method1.tex - David Bailey for Methodology subcommittee (6) lowlev1.tex - Roger Hockney for the low-level subcommittee (7) kernel1.tex - Tony Hey for the kernel subcommittee (8) compac1.tex - David Walker for compact applications subcommittee (9) conclu1.tex - Roger Hockney for whole committee Files (1), (2), (3), (5), (7), and (8) are appended to this e-mail. Best wishes, Roger Hockney. % file : benrep1.tex % % ************************************************************** % STANDARD INTERNATIONAL BENCHMARKS FOR PARALLEL COMPUTERS % ************************************************************** % \input{bencom1.tex} % define new commands for benchmark report % ---------------------------------------------------------------------------- \documentstyle[]{report} % Specifies the document style. \textheight 8.25 true in \textwidth 5.625 true in \topmargin -0.13 true in \oddsidemargin 0.25 true in \evensidemargin 0.25 true in % The preamble begins here. \title{Standard International Benchmarks for Parallel Computers} % ---------------------------------------------------------------------------- \author{PBWG Committee \\ draft assembled by Roger Hockney (chairman)} \date{23 March 1993 - draft 1} % ---------------------------------------------------------------------------- \begin{document} % End of preamble and beginning of text. \sloppy \maketitle % Produces the title. % ---------------------------------------------------------------------------- \input{intro1.tex} % Introduction % responsibility of Roger Hockney for whole committee % ---------------------------------------------------------------------------- \input{method1.tex} % Chapter1 % responsibility of David Bailey for Methodology subcommittee % ---------------------------------------------------------------------------- \input{lowlev1.tex} % Chapter2 % responsibility of Roger Hockney for Low-level benchmarks subcommittee % ---------------------------------------------------------------------------- \input{kernel1.tex} % Chapter3 % responsibility of Tony Hey for Kernel benchmarks subcommittee % ---------------------------------------------------------------------------- \input{compac1.tex} % Chapter4 % responsibility of David Walker for Compact Applications subcommittee % ---------------------------------------------------------------------------- \input{conclu1.tex} % Conclusions % responsibility of Roger Hockney for whole committee % ---------------------------------------------------------------------------- \vspace{0.35in} {\large \bf Acknowledgments} \bibliography{benref1} \bibliographystyle{unsrt} \end{document} % End of document. % file : bencom1.tex % % ************************************************************** % LATEX COMMANDS FOR PARABEN REPORTS % ************************************************************** % \def\flop{\mathop{\rm flop}\nolimits} \def\pipe{\mathop{\rm pipe}\nolimits} \newcommand{\usec}{\mbox{\rm $\mu$s}} \newcommand{\where}{\mbox{\rm where}} \newcommand{\rmand}{\mbox{\rm and}} \newcommand{\Mflops}{\mbox{\rm Mflop/s}} \newcommand{\flops}{\mbox{\rm flop/s}} \newcommand{\flopB}{\mbox{\rm flop/B}} \newcommand{\tstepps}{\mbox{\rm tstep/s}} \newcommand{\MWps}{\mbox{\rm MW/s}} \newcommand{\Mwps}{\mbox{\rm Mw/s}} \newcommand{\spone}{\mbox{\ }} \newcommand{\sptwo}{\mbox{\ \ }} \newcommand{\spfour}{\mbox{\ \ \ \ }} \newcommand{\spsix}{\mbox{\ \ \ \ \ \ }} \newcommand{\speight}{\mbox{\ \ \ \ \ \ \ \ }} \newcommand{\spten}{\mbox{\ \ \ \ \ \ \ \ \ \ }} \newcommand{\rinf}{\mbox{$r_\infty$}} \newcommand{\Rinf}{\mbox{$R_\infty$}} \newcommand{\nhalf}{\mbox{$n_{\frac{1}{2}}$}} \newcommand{\Nhalf}{\mbox{$N_{\frac{1}{2}}$}} \newcommand{\phalf}{\mbox{$p_{\frac{1}{2}}$}} \newcommand{\Phalf}{\mbox{$P_{\frac{1}{2}}$}} \newcommand{\half}{\mbox{$\frac{1}{2}$}} \newcommand{\rnhalf}{\mbox{(\rinf,\nhalf)}} \newcommand{\RNhalf}{\mbox{(\Rinf,\Nhalf)}} \newcommand{\third}{\mbox{$\frac{1}{3}$}} \newcommand{\quart}{\mbox{$\frac{1}{4}$}} \newcommand{\eighth}{\mbox{$\frac{1}{8}$}} \newcommand{\nineth}{\mbox{$\frac{1}{9}$}} % ---------------------------------------------------------------------------- % file : benref1.bib % @book{HoJe88, author= "Roger W. Hockney and Christopher R. Jesshope", title= "Parallel Computers 2: Architecture, Programming and Algorithms", publisher= "Adam Hilger/IOP Publishing", address= "Bristol \& New York", year= "1988", note= "Distributed in the USA by the American Institute of Physics, c/o AIDC, 64 Depot Road, Colchester, VT 05445."} @book{Super, key="Super", title={Supercomputer}, publisher="ASFRA", address="Edam, Netherlands"} @book{SI75, key="Royal Society", organization="Symbols Committee of the Royal Society", title={Quantities, Units and Symbols}, publisher="The Royal Society", address="London", year=1975} @article{Berr89, author="M. Berry and D. Chen and P. Koss and D. Kuck and S. Lo and Y. Pang and L. Pointer and R. Roloff and A. Sameh and E. Clementi and S. Chin and D. Schneider and G. Fox and P. Messina and D. Walker and C. Hsiung and J. Schwarzmeier and K. Lue and S. Orszag and F. Seidl and O. Johnson and R. Goodrum and J. Martin", title="The {PERFECT} Club benchmarks: effective performance evaluation of computers", journal={Intl. J. Supercomputer Appls.}, volume=3, number=3, year=1989, pages="5-40"} @incollection{Ma88, author="F. H. McMahon", title="The {L}ivermore {F}ortran Kernels test of the numerical performance range", editor="J. L. Martin", booktitle={Performance Evaluation of Supercomputers}, publisher="Elsevier Science B.V., North-Holland", address="Amsterdam", year=1988, pages="143-186"} @article{Mess90, author="P. Messina and C. Baillie and E. Felten and P. Hipes and R. Williams and A. Alagar and A. Kamrath and R. Leary and W. Pfeiffer and J. Rogers and D. Walker", title="Benchmarking advanced architecture computers", journal={Concurrency: Practice and Experience}, volume=2, number=3, year=1990, pages="195-255"} @inproceedings{Cvet90, author="Z. Cvetanovic and E. G. Freedman and C. Nofsinger", title="Efficient decomposition and performance of parallel {PDE}, {FFT}, {M}onte-{C}arlo simulations, simplex and sparse solvers", booktitle={Proceedings Supercomputing90}, publisher="IEEE", address="New York", year=1990, pages="465-474"} @article{SUPR88, title="Proceedings 2nd International SUPRENUM Colloquium", author="U. Trottenberg", journal={Parallel Computing}, volume=7, number=3, year=1988} @article{Hey91, author="A. J. G. Hey", title="The {G}enesis Distributed-Memory Benchmarks", journal={Parallel Computing}, volume=17, year=1991, pages="1275-1283"} @book{F90, author="M. Metcalf and J. Reid", title={Fortran-90 Explained}, publisher="Oxford Science Publications/OUP", address="Oxford and New York", year=1990, chapter=6} @article{SPEC90, key="SPEC", title="{SPEC} Benchmarks Suite Release 1.0", journal={SPEC Newslett.}, volume=2, number=3, year=1990, pages="3-4", publisher="Systems Performance Evaluation Cooperative, Waterside Associates", address="Fremont, California"} @article{FGHS89, author="A. Friedli and W. Gentzsch and R. Hockney and A. van der Steen", title="A {E}uropean Supercomputer Benchmark Effort", journal={Supercomputer 34}, volume="VI", number=6, year=1989, pages="14-17"} @article{BRH90, author="L. Bomans and D. Roose and R. Hempel", title="The {A}rgonne/{GMD} Macros in {F}ortran for portable parallel programming and their implementation on the {I}ntel i{PSC}/2", journal={Parallel Computing}, volume=15, year=1990, pages="119-132"} @inproceedings{ShTu91, author="J. N. Shahid and R. S. Tuminaro", title="Iterative Methods for Nonsymmetric Systems on {MIMD} Machines", booktitle={Proc. Fifth SIAM Conf. Parallel Processing for Scientific Computing}, year=1991} @article{Bish90, author="N. T. Bishop and C. J. S. Clarke and R. A. d'Inverno", journal={Classical and Quantum Gravity}, volume=7, year=1990, pages="L23-L27"} @article{Isaac83, author="R. A. Isaacson and J. S. Welling and J.Winicour", journal={J. Math. Phys.}, volume=24, year=1983, pages="1824-1834"} @article{Stew82, author="J. M. Stewart and H. Friedrich", journal={Proc. Roy. Soc.}, volume="A384", year=1982, pages="427-454"} @article{Hoc92, author="R. W. Hockney", title="A framework for benchmark analysis", journal={Supercomputer}, volume=48, number="IX-2", year=1992, pages="9-22"} @article{Add93, author="C. Addison and J. Allwright and N. Binsted and N. Bishop and B. Carpenter and P. Dalloz and D. Gee and V. Getov and A. Hey and R. Hockney and M. Lemke and J. Merlin and M. Pinches and C. Scott and I. Wolton", title="The {G}enesis distributed-memory benchmarks. Part 1: methodology and general relativity benchmark with results for the {SUPRENUM} computer", journal={Concurrency: Practice and Experience}, volume=5, number=1, year=1993, pages="1-22"} % % ------------------------------------------------------------------------- %file: intro1.tex \chapter{Introduction}\footnotemark \footnotetext{written by Roger Hockney for whole committee} \cite{HoJe88} % % ------------------------------------------------------------------------- %file: method1.tex \chapter{Methodology} \footnote{assembled by David Bailey for Methodology subcommittee} % % ------------------------------------------------------------------------- %file: lowlev1.tex \chapter{Low-Level Benchmarks} \footnote{assembled by Roger Hockney for low-level subcommittee} % % ------------------------------------------------------------------------- %file: kernel1.tex \chapter{Kernel Benchmarks} \footnote{assembled by Tony Hey for Kernel subcommittee} % % ------------------------------------------------------------------------- %file: compac1.tex \chapter{Compact Applications} \footnote{assembled by David Walker for Compact Applications subcommittee} % % ------------------------------------------------------------------------- %file: conclu1.tex \chapter{Conclusions} \footnote{written by Roger Hockney for whole committee} % % ------------------------------------------------------------------------- % End of skeleton paper From owner-pbwg-comm@CS.UTK.EDU Mon Mar 29 12:54:46 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA19359; Mon, 29 Mar 93 12:54:46 -0500 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA04636; Mon, 29 Mar 93 12:52:54 -0500 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 29 Mar 1993 12:52:53 EST Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from gemini.npac.syr.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA04626; Mon, 29 Mar 93 12:52:42 -0500 Received: from localhost by gemini.npac.syr.edu with SMTP id AA11964 (5.65c/IDA-1.4.4 for pbwg-comm@cs.utk.edu); Mon, 29 Mar 1993 12:52:02 -0500 Message-Id: <199303291752.AA11964@gemini.npac.syr.edu> To: Prof Roger Hockney Cc: pbwg-comm@cs.utk.edu, haupt@npac.syr.edu Subject: Re: PBWG Draft Report In-Reply-To: Your message of "Wed, 24 Mar 93 12:34:10 GMT." <253.9303241234@alithea.ecs.soton.ac.uk> Date: Mon, 29 Mar 93 12:51:54 -0500 From: haupt@npac.syr.edu X-Mts: smtp I thought we agreed to add additional section to the draft on compiler benchmark. Syracuse volunteers to coordinate that effort, and to provide the text for the draft. Will you add this section to the draft, please? Tom Haupt From owner-pbwg-comm@CS.UTK.EDU Tue Apr 27 03:01:42 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA29378; Tue, 27 Apr 93 03:01:42 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA20445; Tue, 27 Apr 93 03:00:40 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 27 Apr 1993 03:00:39 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from swiba9.unibas.ch by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA20392; Tue, 27 Apr 93 03:00:32 -0400 Received: from sir.ifi.unibas.ch by swiba9.unibas.ch with SMTP (PP) id <6953-0@swiba9.unibas.ch>; Tue, 27 Apr 1993 09:00:04 +0200 Received: from charlie by sir.ifi.unibas.ch (NX5.67c/NX3.0M) id AA07530; Tue, 27 Apr 93 08:59:58 +0200 From: (Walter Kuhn) kuhn@ifi.unibas.ch Message-Id: <9304270659.AA07530@sir.ifi.unibas.ch> Received: by charlie.ifi.unibas.ch (NX5.67c/NX3.0X) id AA00297; Tue, 27 Apr 93 08:59:58 +0200 Date: Tue, 27 Apr 93 08:59:58 +0200 To: pbwg-comm@CS.UTK.edu Subject: Additions to pbwg maillist Please add me to your mail list. Walter From owner-pbwg-comm@CS.UTK.EDU Fri Apr 30 12:32:36 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA18590; Fri, 30 Apr 93 12:32:36 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA22997; Fri, 30 Apr 93 12:31:04 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Fri, 30 Apr 1993 12:30:51 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA22949; Fri, 30 Apr 93 12:30:47 -0400 Via: uk.ac.southampton.ecs; Fri, 30 Apr 1993 16:11:19 +0100 From: R.Hockney@parallel-applications-centre.southampton.ac.uk Via: calvados.pac.soton.ac.uk (plonk); Fri, 30 Apr 93 16:03:41 BST Date: Fri, 30 Apr 93 15:10:51 GMT Message-Id: <15929.9304301510@calvados.pac.soton.ac.uk> To: pbwg-comm@cs.utk.edu Subject: Revised Report Framework Updated Report Framework and Bibliography ----------------------------------------- I append to this note updated versions of benrep1.tex and benref1.bib, there is an added chapter on Compiler benchmarks and additional references. I have sent a contribution to Chapter 2 to pbwg-method and a draft of Chapter 3 to pbwg-lowlevel. I await drafts of the other chapters from the different subcommittee leaders: Roger Hockney ------------------------------- cut here -------------------------------- % % ************************************************************** % STANDARD INTERNATIONAL BENCHMARKS FOR PARALLEL COMPUTERS % ************************************************************** % \input{bencom1.tex} % define new commands for benchmark report % ---------------------------------------------------------------------------- \documentstyle[]{report} % Specifies the document style. \textheight 8.25 true in \textwidth 5.625 true in \topmargin -0.13 true in \oddsidemargin 0.25 true in \evensidemargin 0.25 true in % The preamble begins here. \title{Standard International Benchmarks for Parallel Computers} % ---------------------------------------------------------------------------- \author{PBWG Committee \\ draft assembled by Roger Hockney (chairman)} \date{19 April 1993 - draft 2} % ---------------------------------------------------------------------------- \begin{document} % End of preamble and beginning of text. \sloppy \maketitle % Produces the title. % ---------------------------------------------------------------------------- \input{intro1.tex} % Introduction % responsibility of Roger Hockney for whole committee % ---------------------------------------------------------------------------- \input{method3.tex} % Chapter1 % responsibility of David Bailey for Methodology subcommittee % ---------------------------------------------------------------------------- \input{lowlev1.tex} % Chapter2 % responsibility of Roger Hockney for Low-level benchmarks subcommittee % ---------------------------------------------------------------------------- \input{kernel1.tex} % Chapter3 % responsibility of Tony Hey for Kernel benchmarks subcommittee % ---------------------------------------------------------------------------- \input{compac1.tex} % Chapter4 % responsibility of David Walker for Compact Applications subcommittee % ---------------------------------------------------------------------------- \input{compil1.tex} % Chapter5 % responsibility of Tom Haupt for Compiler Benchmarks subcommittee % ---------------------------------------------------------------------------- \input{conclu1.tex} % Conclusions % responsibility of Roger Hockney for whole committee % ---------------------------------------------------------------------------- \vspace{0.35in} {\large \bf Acknowledgments} \bibliography{benref1} \bibliographystyle{unsrt} \end{document} % End of document. % @book{HoJe81, author= "Roger W. Hockney and Christopher R. Jesshope", title= "Parallel Computers: Architecture, Programming and Algorithms", publisher= "Adam Hilger", address= "Bristol", year= "1981", } @book{HoJe88, author= "Roger W. Hockney and Christopher R. Jesshope", title= "Parallel Computers 2: Architecture, Programming and Algorithms", publisher= "Adam Hilger/IOP Publishing", address= "Bristol \& Philadelphia", year= "1988", edition="second", note= "Distributed in the USA by IOP Publ. Inc., Public Ledger Bldg., Suite 1035, Independence Square, Philadelphia, PA 19106."} @book{Super, key="Super", title={Supercomputer}, publisher="ASFRA", address="Edam, Netherlands"} @book{SI75, key="Royal Society", organization="Symbols Committee of the Royal Society", title={Quantities, Units and Symbols}, publisher="The Royal Society", address="London", year=1975} @article{Berr89, author="M. Berry and D. Chen and P. Koss and D. Kuck and S. Lo and Y. Pang and L. Pointer and R. Roloff and A. Sameh and E. Clementi and S. Chin and D. Schneider and G. Fox and P. Messina and D. Walker and C. Hsiung and J. Schwarzmeier and K. Lue and S. Orszag and F. Seidl and O. Johnson and R. Goodrum and J. Martin", title="The {PERFECT} Club benchmarks: effective performance evaluation of computers", journal={Intl. J. Supercomputer Appls.}, volume=3, number=3, year=1989, pages="5-40"} @incollection{Ma88, author="F. H. McMahon", title="The {L}ivermore {F}ortran {K}ernels test of the numerical performance range", editor="J. L. Martin", booktitle={Performance Evaluation of Supercomputers}, publisher="Elsevier Science B.V., North-Holland", address="Amsterdam", year=1988, pages="143-186"} @article{Mess90, author="P. Messina and C. Baillie and E. Felten and P. Hipes and R. Williams and A. Alagar and A. Kamrath and R. Leary and W. Pfeiffer and J. Rogers and D. Walker", title="Benchmarking advanced architecture computers", journal={Concurrency: Practice and Experience}, volume=2, number=3, year=1990, pages="195-255"} @inproceedings{Cvet90, author="Z. Cvetanovic and E. G. Freedman and C. Nofsinger", title="Efficient decomposition and performance of parallel {PDE}, {FFT}, {M}onte-{C}arlo simulations, simplex and sparse solvers", booktitle={Proceedings Supercomputing90}, publisher="IEEE", address="New York", year=1990, pages="465-474"} @article{SUPR88, title="Proceedings 2nd International SUPRENUM Colloquium", author="U. Trottenberg", journal={Parallel Computing}, volume=7, number=3, year=1988} @article{Hey91, author="A. J. G. Hey", title="The {G}enesis Distributed-Memory Benchmarks", journal={Parallel Computing}, volume=17, year=1991, pages="1275-1283"} @book{F90, author="M. Metcalf and J. Reid", title={Fortran-90 Explained}, publisher="Oxford Science Publications/OUP", address="Oxford and New York", year=1990, chapter=6} @article{SPEC90, key="SPEC", title="{SPEC} Benchmarks Suite Release 1.0", journal={SPEC Newslett.}, volume=2, number=3, year=1990, pages="3-4", publisher="Systems Performance Evaluation Cooperative, Waterside Associates", address="Fremont, California"} @article{FGHS89, author="A. Friedli and W. Gentzsch and R. Hockney and A. van der Steen", title="A {E}uropean Supercomputer Benchmark Effort", journal={Supercomputer 34}, volume="VI", number=6, year=1989, pages="14-17"} @article{BRH90, author="L. Bomans and D. Roose and R. Hempel", title="The {A}rgonne/{GMD} Macros in {F}ortran for portable parallel programming and their implementation on the {I}ntel i{PSC}/2", journal={Parallel Computing}, volume=15, year=1990, pages="119-132"} @inproceedings{ShTu91, author="J. N. Shahid and R. S. Tuminaro", title="Iterative Methods for Nonsymmetric Systems on {MIMD} Machines", booktitle={Proc. Fifth SIAM Conf. Parallel Processing for Scientific Computing}, year=1991} @article{Bish90, author="N. T. Bishop and C. J. S. Clarke and R. A. d'Inverno", journal={Classical and Quantum Gravity}, volume=7, year=1990, pages="L23-L27"} @article{Isaac83, author="R. A. Isaacson and J. S. Welling and J.Winicour", journal={J. Math. Phys.}, volume=24, year=1983, pages="1824-1834"} @article{Stew82, author="J. M. Stewart and H. Friedrich", journal={Proc. Roy. Soc.}, volume="A384", year=1982, pages="427-454"} @incollection{Hoc77, author="R. W. Hockney", title="Super-Computer Architecture", editor="F. Sumner", booktitle={Infotech State of the Art Conference: {F}uture {S}ystems}, publisher="Infotech", address="Maidenhead", year=1977, pages="277-305"} @article{Hoc82, author="R. W. Hockney", title="Characterization of parallel computers and algorithms", journal={Computer Physics Communications}, volume=26, year=1982, pages="285-291"} @article{Hoc83, author="R. W. Hockney", title="Characterizing Computers and Optimizing the {FACR}(l) Poisson-Solver on Parallel Unicomputers", journal={IEEE Trans. Comput.}, volume="{C}\-32", year=1983, pages="933-941"} @article{Hoc87, author="R. W. Hockney", title="Parametrization of Computer Performance", journal={Parallel Computing}, volume=5, year=1987, pages="97-103"} @article{Hoc88, author="R. W. Hockney", title="Synchronization and Communication Overheads on the {LCAP} Multiple {FPS}-164 Computer System", journal={Parallel Computing}, volume=9, year=1988, pages="279-290"} @article{HoCu89, author="R. W. Hockney and I. J. Curington", title="$f_{frac{1}{2}}$: a Parameter to Characterise Memory and Communication Bottlenecks", journal={Parallel Computing}, volume=10, year=1989, pages="277-286"} @article{Hoc91, author="R. W. Hockney", title="Performance Parameters and Benchmarking of Supercomputers", journal={Parallel Computing}, volume=17, year=1991, pages="1111-1130"} @article{Hoc92, author="R. W. Hockney", title="A framework for benchmark analysis", journal={Supercomputer}, volume=48, number="IX-2", year=1992, pages="9-22"} @article{HoCa92, author="R. W. Hockney and E. A. Carmona", title="Comparison of Communications on the {I}ntel i{PSC}/860 and {T}ouchstone {D}elta", journal={Parallel Computing}, volume=18, year=1992, pages="1067-1072"} @article{Add93, author="C. Addison and J. Allwright and N. Binsted and N. Bishop and B. Carpenter and P. Dalloz and D. Gee and V. Getov and A. Hey and R. Hockney and M. Lemke and J. Merlin and M. Pinches and C. Scott and I. Wolton", title="The {G}enesis distributed-memory benchmarks. Part 1: methodology and general relativity benchmark with results for the {SUPRENUM} computer", journal={Concurrency: Practice and Experience}, volume=5, number=1, year=1993, pages="1-22"} @techreport{StRi93, author="A. J. van der Steen and P. P. M. de Rijk", title="Guidelines for use of the {E}uro{B}en Benchmark", institution="{E}uro{B}en", year=1993, month=feb, type="Technical Report", number="{TR}\-3", address="The {E}uro{B}en Group, {U}trecht, {T}he {N}etherlands"} From owner-pbwg-comm@CS.UTK.EDU Fri Apr 30 12:49:20 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA20526; Fri, 30 Apr 93 12:49:20 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA24215; Fri, 30 Apr 93 12:48:17 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Fri, 30 Apr 1993 12:48:16 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from vnet.ibm.com by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA24207; Fri, 30 Apr 93 12:48:15 -0400 Message-Id: <9304301648.AA24207@CS.UTK.EDU> Received: from KGNVMZ by vnet.IBM.COM (IBM VM SMTP V2R2) with BSMTP id 8031; Fri, 30 Apr 93 12:48:11 EDT Date: Fri, 30 Apr 93 12:47:34 EDT From: "Dr. Joanne L. Martin ((914) 385-9572)" To: pbwg-comm@cs.utk.edu Subject: New address Please note that my new e-mail address is jmartin at vnet.ibm.com Thanks, Joanne. From owner-pbwg-comm@CS.UTK.EDU Mon May 3 11:05:02 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA17408; Mon, 3 May 93 11:05:02 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02905; Mon, 3 May 93 11:03:37 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 3 May 1993 11:03:35 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from swiba9.unibas.ch by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02897; Mon, 3 May 93 11:03:28 -0400 Received: from sir.ifi.unibas.ch by swiba9.unibas.ch with SMTP (PP) id <21115-0@swiba9.unibas.ch>; Mon, 3 May 1993 17:03:12 +0200 Received: from charlie by sir.ifi.unibas.ch (NX5.67c/NX3.0M) id AA27670; Mon, 3 May 93 17:03:07 +0200 From: (Walter Kuhn) kuhn@ifi.unibas.ch Message-Id: <9305031503.AA27670@sir.ifi.unibas.ch> Received: by charlie.ifi.unibas.ch (NX5.67c/NX3.0X) id AA00509; Mon, 3 May 93 17:03:05 +0200 Date: Mon, 3 May 93 17:03:05 +0200 Received: by NeXT.Mailer (1.87.1) Received: by NeXT Mailer (1.87.1) To: pbwg-comm@cs.utk.edu Subject: send lowlevel.archive from pbwg From owner-pbwg-comm@CS.UTK.EDU Mon May 3 11:05:13 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA17427; Mon, 3 May 93 11:05:13 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02925; Mon, 3 May 93 11:04:06 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 3 May 1993 11:04:05 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from swiba9.unibas.ch by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02909; Mon, 3 May 93 11:03:56 -0400 Received: from sir.ifi.unibas.ch by swiba9.unibas.ch with SMTP (PP) id <21120-0@swiba9.unibas.ch>; Mon, 3 May 1993 17:03:33 +0200 Received: from charlie by sir.ifi.unibas.ch (NX5.67c/NX3.0M) id AA27674; Mon, 3 May 93 17:03:30 +0200 From: (Walter Kuhn) kuhn@ifi.unibas.ch Message-Id: <9305031503.AA27674@sir.ifi.unibas.ch> Received: by charlie.ifi.unibas.ch (NX5.67c/NX3.0X) id AA00514; Mon, 3 May 93 17:03:30 +0200 Date: Mon, 3 May 93 17:03:30 +0200 Received: by NeXT.Mailer (1.87.1) Received: by NeXT Mailer (1.87.1) To: pbwg-comm@cs.utk.edu Subject: send method.archive from pbwg From owner-pbwg-comm@CS.UTK.EDU Mon May 3 11:05:38 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA17445; Mon, 3 May 93 11:05:38 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02939; Mon, 3 May 93 11:04:35 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 3 May 1993 11:04:33 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from swiba9.unibas.ch by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02928; Mon, 3 May 93 11:04:24 -0400 Received: from sir.ifi.unibas.ch by swiba9.unibas.ch with SMTP (PP) id <21132-0@swiba9.unibas.ch>; Mon, 3 May 1993 17:04:14 +0200 Received: from charlie by sir.ifi.unibas.ch (NX5.67c/NX3.0M) id AA27678; Mon, 3 May 93 17:04:10 +0200 From: (Walter Kuhn) kuhn@ifi.unibas.ch Message-Id: <9305031504.AA27678@sir.ifi.unibas.ch> Received: by charlie.ifi.unibas.ch (NX5.67c/NX3.0X) id AA00519; Mon, 3 May 93 17:04:10 +0200 Date: Mon, 3 May 93 17:04:10 +0200 Received: by NeXT.Mailer (1.87.1) Received: by NeXT Mailer (1.87.1) To: pbwg-comm@cs.utk.edu Subject: send compil.archive from pbwg From owner-pbwg-comm@CS.UTK.EDU Mon May 3 11:05:58 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA17467; Mon, 3 May 93 11:05:58 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02984; Mon, 3 May 93 11:05:07 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 3 May 1993 11:05:05 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from swiba9.unibas.ch by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02948; Mon, 3 May 93 11:04:53 -0400 Received: from sir.ifi.unibas.ch by swiba9.unibas.ch with SMTP (PP) id <21137-0@swiba9.unibas.ch>; Mon, 3 May 1993 17:04:34 +0200 Received: from charlie by sir.ifi.unibas.ch (NX5.67c/NX3.0M) id AA27683; Mon, 3 May 93 17:04:30 +0200 From: (Walter Kuhn) kuhn@ifi.unibas.ch Message-Id: <9305031504.AA27683@sir.ifi.unibas.ch> Received: by charlie.ifi.unibas.ch (NX5.67c/NX3.0X) id AA00525; Mon, 3 May 93 17:04:30 +0200 Date: Mon, 3 May 93 17:04:30 +0200 Received: by NeXT.Mailer (1.87.1) Received: by NeXT Mailer (1.87.1) To: pbwg-comm@cs.utk.edu Subject: send conclu.archive from pbwg From owner-pbwg-comm@CS.UTK.EDU Wed May 5 09:08:29 1993 Received: from CS.UTK.EDU by surfer.EPM.ORNL.GOV (5.61/1.34) id AA15181; Wed, 5 May 93 09:08:29 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA05335; Wed, 5 May 93 06:47:02 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 5 May 1993 06:46:59 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from THUD.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA05327; Wed, 5 May 93 06:46:53 -0400 From: Jack Dongarra Received: by thud.cs.utk.edu (5.61+IDA+UTK-930125/2.7c-UTK) id AA08869; Wed, 5 May 93 06:46:47 -0400 Date: Wed, 5 May 93 06:46:47 -0400 Message-Id: <9305051046.AA08869@thud.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: PBWG Meeting May 24th Dear Colleague, We are planning to have the Third Meeting of the Parallel Benchmark Working Group meet in Knoxville, Tennessee at the University of Tennessee on May 24th, 1993. This process formally began with a workshop held at the Supercomputer '92 meeting in November 1992. The purpose of the working group is to establish credible and useful benchmarks for the evaluation of Distributed Memory MIMD systems. The objectives for the group are: 1. To establish a comprehensive set of parallel benchmarks that is generally accepted by both users and vendors of parallel system. 2. To provide a focus for parallel benchmark activities and avoid unnecessary duplication of effort and proliferation of benchmarks. 3. To set standards for benchmarking methodology and result-reporting together with a control database/repository for both the benchmarks and the results. Mode of Working: The working group has adopted an HPF-like forum style of proceedings with a view to convergence to an agreed set of benchmarks and procedures within 10 months. If you would like to participate and attend the meeting let me know. Mailing Lists ============= The following mailing lists have been set up. pbwg-comm@cs.utk.edu Whole committee pbwg-lowlevel@cs.utk.edu Low level subcommittee pbwg-compactapp@cs.utk.edu Compact applications subcommittee pbwg-method@cs.utk.edu Methodology subcommittee pbwg-kernel@cs.utk.edu Kernel subcommittee If you are on a mailing list you will receive mail as it is posted. If you want to join a mailing list send me mail (dongarra@cs.utk.edu). All mail will be collected and can be retrieved by sending email to netlib@ornl.gov and in the mail message typing: send comm.archive from pbwg send lowlevel.archive from pbwg send compactapp.archive from pbwg send method.archive from pbwg send kernel.archive from pbwg send index from pbwg The various subcommittees will look into the following topics: Low-Level: --------- Start-up, latency, bandwidth Reduction (broadcast, sum, gather/scatter) Synchronization (e.g., SYNCH1 from Genesis) I/O Kernel: ------ Matrix operations (e.g., multiply, transpose) LU Decomposition PDE Solvers (Red/Black Relaxation) Multigrid FFT Conjugate Gradient Compact Applications: -------------------- Particle-In-Cell codes (e.g., LPM1 from Genesis) QCD Molecular Dynamics CFD ARCO Financial Applications Methodology: ------------ Guidelines for reporting performance. The meeting site will be the: Science Alliance Conference Room South College University of Tennessee (A postscript map in included at the end of this message, South College is the building located next to Ayres Hall.) We have made arrangements with the Hilton Hotel in Knoxville. Hilton Hotel 501 W. Church Street Knoxville, TN Phone: 615-523-2300 When making arrangements tell the hotel you are associated with the Parallel Benchmarking Meeting. The rate is $65.00/night. You can rent a car or get a cab from the airport to the hotel. From the hotel to the University it is a 15 minute walk. We should plan to start at 8:30 am May 24th and finish about 5:00 pm. The format of the meeting is: Monday 24th May 8.30 - 12.00 Full group meeting 12.00 - 1.30 Lunch 1.30 - 4.00 Parallel subgroup meetings 4.00 - 5.00 Full group meeting Tentative agenda for full group meeting: 1. Minutes of Minneapolis meeting 2. Reports and discussion from subgroups 3. Open discussion and agreement on further actions 4. Date and venue for next meeting Suggested subgroups - probably two in parallel Compact Applications Low-Level benchmarks and second pair: Kernels benchmarks Methodology We have setup a mail refector for correspondence, it is called pbwg-comm@cs.utk.edu. Mail to that address will be sent to the mailing list and also collected in netlib@ornl.gov. To retrieve the collected mail, send email to netlib@ornl.gov and in the mail message type: send comm.archive from pbwg If you would like to be put on the mailing list to receive the correspondence let me know. Regards, Jack Dongarra ---postscript map of the University of Tennessee--- %!PS-Adobe-2.0 EPSF-1.2 %%DocumentFonts: Helvetica-Bold Courier Courier-Bold Times-Bold %%Pages: 1 %%BoundingBox: 39 -113 604 767 %%EndComments /arrowHeight 10 def /arrowWidth 5 def /IdrawDict 54 dict def IdrawDict begin /reencodeISO { dup dup findfont dup length dict begin { 1 index /FID ne { def }{ pop pop } ifelse } forall /Encoding ISOLatin1Encoding def currentdict end definefont } def /ISOLatin1Encoding [ /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright /parenleft/parenright/asterisk/plus/comma/minus/period/slash /zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon /less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N /O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright /asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m /n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef /.notdef/dotlessi/grave/acute/circumflex/tilde/macron/breve /dotaccent/dieresis/.notdef/ring/cedilla/.notdef/hungarumlaut /ogonek/caron/space/exclamdown/cent/sterling/currency/yen/brokenbar /section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot /hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior /acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine /guillemotright/onequarter/onehalf/threequarters/questiondown /Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla /Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex /Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis /multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute /Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis /aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave /iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex /otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis /yacute/thorn/ydieresis ] def /Helvetica-Bold reencodeISO def /Courier reencodeISO def /Courier-Bold reencodeISO def /Times-Bold reencodeISO def /none null def /numGraphicParameters 17 def /stringLimit 65535 def /Begin { save numGraphicParameters dict begin } def /End { end restore } def /SetB { dup type /nulltype eq { pop false /brushRightArrow idef false /brushLeftArrow idef true /brushNone idef } { /brushDashOffset idef /brushDashArray idef 0 ne /brushRightArrow idef 0 ne /brushLeftArrow idef /brushWidth idef false /brushNone idef } ifelse } def /SetCFg { /fgblue idef /fggreen idef /fgred idef } def /SetCBg { /bgblue idef /bggreen idef /bgred idef } def /SetF { /printSize idef /printFont idef } def /SetP { dup type /nulltype eq { pop true /patternNone idef } { dup -1 eq { /patternGrayLevel idef /patternString idef } { /patternGrayLevel idef } ifelse false /patternNone idef } ifelse } def /BSpl { 0 begin storexyn newpath n 1 gt { 0 0 0 0 0 0 1 1 true subspline n 2 gt { 0 0 0 0 1 1 2 2 false subspline 1 1 n 3 sub { /i exch def i 1 sub dup i dup i 1 add dup i 2 add dup false subspline } for n 3 sub dup n 2 sub dup n 1 sub dup 2 copy false subspline } if n 2 sub dup n 1 sub dup 2 copy 2 copy false subspline patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if brushNone not { istroke } if 0 0 1 1 leftarrow n 2 sub dup n 1 sub dup rightarrow } if end } dup 0 4 dict put def /Circ { newpath 0 360 arc patternNone not { ifill } if brushNone not { istroke } if } def /CBSpl { 0 begin dup 2 gt { storexyn newpath n 1 sub dup 0 0 1 1 2 2 true subspline 1 1 n 3 sub { /i exch def i 1 sub dup i dup i 1 add dup i 2 add dup false subspline } for n 3 sub dup n 2 sub dup n 1 sub dup 0 0 false subspline n 2 sub dup n 1 sub dup 0 0 1 1 false subspline patternNone not { ifill } if brushNone not { istroke } if } { Poly } ifelse end } dup 0 4 dict put def /Elli { 0 begin newpath 4 2 roll translate scale 0 0 1 0 360 arc patternNone not { ifill } if brushNone not { istroke } if end } dup 0 1 dict put def /Line { 0 begin 2 storexyn newpath x 0 get y 0 get moveto x 1 get y 1 get lineto brushNone not { istroke } if 0 0 1 1 leftarrow 0 0 1 1 rightarrow end } dup 0 4 dict put def /MLine { 0 begin storexyn newpath n 1 gt { x 0 get y 0 get moveto 1 1 n 1 sub { /i exch def x i get y i get lineto } for patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if brushNone not { istroke } if 0 0 1 1 leftarrow n 2 sub dup n 1 sub dup rightarrow } if end } dup 0 4 dict put def /Poly { 3 1 roll newpath moveto -1 add { lineto } repeat closepath patternNone not { ifill } if brushNone not { istroke } if } def /Rect { 0 begin /t exch def /r exch def /b exch def /l exch def newpath l b moveto l t lineto r t lineto r b lineto closepath patternNone not { ifill } if brushNone not { istroke } if end } dup 0 4 dict put def /Text { ishow } def /idef { dup where { pop pop pop } { exch def } ifelse } def /ifill { 0 begin gsave patternGrayLevel -1 ne { fgred bgred fgred sub patternGrayLevel mul add fggreen bggreen fggreen sub patternGrayLevel mul add fgblue bgblue fgblue sub patternGrayLevel mul add setrgbcolor eofill } { eoclip originalCTM setmatrix pathbbox /t exch def /r exch def /b exch def /l exch def /w r l sub ceiling cvi def /h t b sub ceiling cvi def /imageByteWidth w 8 div ceiling cvi def /imageHeight h def bgred bggreen bgblue setrgbcolor eofill fgred fggreen fgblue setrgbcolor w 0 gt h 0 gt and { l b translate w h scale w h true [w 0 0 h neg 0 h] { patternproc } imagemask } if } ifelse grestore end } dup 0 8 dict put def /istroke { gsave brushDashOffset -1 eq { [] 0 setdash 1 setgray } { brushDashArray brushDashOffset setdash fgred fggreen fgblue setrgbcolor } ifelse brushWidth setlinewidth originalCTM setmatrix stroke grestore } def /ishow { 0 begin gsave fgred fggreen fgblue setrgbcolor /fontDict printFont printSize scalefont dup setfont def /descender fontDict begin 0 [FontBBox] 1 get FontMatrix end transform exch pop def /vertoffset 1 printSize sub descender sub def { 0 vertoffset moveto show /vertoffset vertoffset printSize sub def } forall grestore end } dup 0 3 dict put def /patternproc { 0 begin /patternByteLength patternString length def /patternHeight patternByteLength 8 mul sqrt cvi def /patternWidth patternHeight def /patternByteWidth patternWidth 8 idiv def /imageByteMaxLength imageByteWidth imageHeight mul stringLimit patternByteWidth sub min def /imageMaxHeight imageByteMaxLength imageByteWidth idiv patternHeight idiv patternHeight mul patternHeight max def /imageHeight imageHeight imageMaxHeight sub store /imageString imageByteWidth imageMaxHeight mul patternByteWidth add string def 0 1 imageMaxHeight 1 sub { /y exch def /patternRow y patternByteWidth mul patternByteLength mod def /patternRowString patternString patternRow patternByteWidth getinterval def /imageRow y imageByteWidth mul def 0 patternByteWidth imageByteWidth 1 sub { /x exch def imageString imageRow x add patternRowString putinterval } for } for imageString end } dup 0 12 dict put def /min { dup 3 2 roll dup 4 3 roll lt { exch } if pop } def /max { dup 3 2 roll dup 4 3 roll gt { exch } if pop } def /midpoint { 0 begin /y1 exch def /x1 exch def /y0 exch def /x0 exch def x0 x1 add 2 div y0 y1 add 2 div end } dup 0 4 dict put def /thirdpoint { 0 begin /y1 exch def /x1 exch def /y0 exch def /x0 exch def x0 2 mul x1 add 3 div y0 2 mul y1 add 3 div end } dup 0 4 dict put def /subspline { 0 begin /movetoNeeded exch def y exch get /y3 exch def x exch get /x3 exch def y exch get /y2 exch def x exch get /x2 exch def y exch get /y1 exch def x exch get /x1 exch def y exch get /y0 exch def x exch get /x0 exch def x1 y1 x2 y2 thirdpoint /p1y exch def /p1x exch def x2 y2 x1 y1 thirdpoint /p2y exch def /p2x exch def x1 y1 x0 y0 thirdpoint p1x p1y midpoint /p0y exch def /p0x exch def x2 y2 x3 y3 thirdpoint p2x p2y midpoint /p3y exch def /p3x exch def movetoNeeded { p0x p0y moveto } if p1x p1y p2x p2y p3x p3y curveto end } dup 0 17 dict put def /storexyn { /n exch def /y n array def /x n array def n 1 sub -1 0 { /i exch def y i 3 2 roll put x i 3 2 roll put } for } def %%EndProlog %%BeginIdrawPrologue /arrowhead { 0 begin transform originalCTM itransform /taily exch def /tailx exch def transform originalCTM itransform /tipy exch def /tipx exch def /dy tipy taily sub def /dx tipx tailx sub def /angle dx 0 ne dy 0 ne or { dy dx atan } { 90 } ifelse def gsave originalCTM setmatrix tipx tipy translate angle rotate newpath arrowHeight neg arrowWidth 2 div moveto 0 0 lineto arrowHeight neg arrowWidth 2 div neg lineto patternNone not { originalCTM setmatrix /padtip arrowHeight 2 exp 0.25 arrowWidth 2 exp mul add sqrt brushWidth mul arrowWidth div def /padtail brushWidth 2 div def tipx tipy translate angle rotate padtip 0 translate arrowHeight padtip add padtail add arrowHeight div dup scale arrowheadpath ifill } if brushNone not { originalCTM setmatrix tipx tipy translate angle rotate arrowheadpath istroke } if grestore end } dup 0 9 dict put def /arrowheadpath { newpath arrowHeight neg arrowWidth 2 div moveto 0 0 lineto arrowHeight neg arrowWidth 2 div neg lineto } def /leftarrow { 0 begin y exch get /taily exch def x exch get /tailx exch def y exch get /tipy exch def x exch get /tipx exch def brushLeftArrow { tipx tipy tailx taily arrowhead } if end } dup 0 4 dict put def /rightarrow { 0 begin y exch get /tipy exch def x exch get /tipx exch def y exch get /taily exch def x exch get /tailx exch def brushRightArrow { tipx tipy tailx taily arrowhead } if end } dup 0 4 dict put def %%EndIdrawPrologue %I Idraw 10 Grid 2.84217e-39 0 %%Page: 1 1 Begin %I b u %I cfg u %I cbg u %I f u %I p u %I t [ 0.799705 0 0 0.799705 0 0 ] concat /originalCTM matrix currentmatrix def Begin %I Pict %I b u %I cfg u %I cbg u %I f u %I p u %I t [ 1 0 0 1 89.1002 831.2 ] concat Begin %I Elli %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.09512e-08 0.9 -0.9 1.09512e-08 584.1 0.89999 ] concat %I 79 550 18 17 Elli End Begin %I Line %I b 65535 3 0 1 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.09512e-08 0.9 -0.9 1.09512e-08 541.8 -27 ] concat %I 110 466 110 542 Line %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.09512e-08 0.9 -0.9 1.09512e-08 541.8 -27 ] concat %I 82 504 140 504 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-helvetica-bold-r-*-140-* Helvetica-Bold 14 SetF %I t [ 1.2168e-08 1 -1 1.2168e-08 35.5 66.1 ] concat %I [ (N) ] Text End End %I eop Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.23147 0.0157385 -0.0157385 -1.23147 409.218 -127.169 ] concat %I [ (Voluteer Boulevard) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 3.66661e-08 3.01333 -3.01333 3.66661e-08 57.9267 164.513 ] concat %I [ (UT Campus -- Jack Dongarra's Lab) ] Text End Begin %I Rect none SetB %I b n %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 1 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 693.5 -156 ] concat %I 17 61 177 478 Rect End Begin %I Line %I b 65535 2 1 1 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 6.9812e-09 0.698798 -0.573736 8.50294e-09 453.281 -79.7194 ] concat %I 158 545 1015 545 Line %I 1 End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 333.274 -81.9323 ] concat %I 257 403 4 4 Elli End Begin %I Line %I b 65535 2 0 1 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 7.02094e-09 0.702776 -0.577002 8.55135e-09 455.344 -80.5588 ] concat %I 211 544 211 17 Line %I 1 End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 6.9812e-09 0.698798 -0.573736 8.50294e-09 455.594 -124.448 ] concat %I 5 503 545 514 535 546 529 628 529 628 529 5 BSpl %I 1 End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 333.274 76.6169 ] concat %I 257 403 4 4 Elli End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 403.382 -81.1611 ] concat %I 257 403 4 4 Elli End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.15378e-08 1.1549 -0.948215 1.40528e-08 222.192 481.065 ] concat %I [ (DOWN) (TOWN) ] Text End Begin %I Poly %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.75 SetP %I t [ 2.43189e-09 0.222447 -0.19986 2.70672e-09 310.54 254.841 ] concat %I 4 301 50 850 50 850 537 301 537 4 Poly End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 9 251 541 251 516 251 507 257 495 263 489 413 491 472 496 485 499 485 499 9 BSpl %I 1 End Begin %I BSpl %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.18822e-09 0.43379 -0.755115 5.27834e-09 620.126 111.397 ] concat %I 6 486 498 502 500 510 504 514 514 514 529 513 542 6 BSpl %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2378 ] concat %I 476 654 476 542 Line %I 1 End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 445.797 -81.1611 ] concat %I 257 403 4 4 Elli End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 252.947 159.612 ] concat %I [ (Volunteer Boulevard) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 310.947 152.672 ] concat %I [ (Neyland Drive) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 198.965 155.756 ] concat %I [ (Cumberland Avenue) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 130.329 90.977 ] concat %I [ (Interstate 40) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 130.329 403.305 ] concat %I [ (Interstate 40) ] Text End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0 SetP %I t [ 5.81767e-09 0.582331 -0.478114 7.08579e-09 332.433 366.124 ] concat %I 257 403 4 4 Elli End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 107 542 486 541 Line %I 1 End Begin %I Line %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 487 541 664 541 Line %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 665 541 682 540 Line %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -51.4666 ] concat %I 681 599 681 485 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.931845 2.26773e-08 -2.26773e-08 -0.931845 205.214 462.905 ] concat %I [ (Henley Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.931845 2.26773e-08 -2.26773e-08 -0.931845 193.647 318.042 ] concat %I [ (17th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 110.137 205.729 ] concat %I [ (17th Street Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 110.137 26.0435 ] concat %I [ (Airport/Alcoa Highway Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-times-bold-r-*-140-* Times-Bold 14 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 171.727 80.0262 ] concat %I [ (Cumberland Avenue Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 265.915 92.3651 ] concat %I [ (Neyland Drive Exit) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 110.137 509.574 ] concat %I [ (Summit Hill Exit) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 144 661 155 636 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 372 662 361 634 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2379 ] concat %I 752 660 736 634 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2378 ] concat %I 186 466 160 486 Line %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 9.38371e-09 0.771182 -0.771182 9.38371e-09 628.85 -52.2378 ] concat %I 180 576 157 542 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -5.99997 ] concat %I 475 50 328 492 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -5.99994 ] concat %I 475 483 329 589 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -6 ] concat %I 962 483 450 588 Line %I 1 End Begin %I Line %I b 65520 0 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -6 ] concat %I 450 491 474 471 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 456.136 36.7289 ] concat %I [ (To Airport) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 9.63786e-09 0.792068 -0.792068 9.63786e-09 137.137 639.729 ] concat %I [ (To Ashville, Bristol) ] Text End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ -0.288762 0.966405 -0.966405 -0.288762 1070.92 66.5381 ] concat %I 743 193 51 70 Elli End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 600.094 464.292 ] concat %I 11 203 116 212 116 212 140 226 140 226 122 222 122 222 102 212 102 212 110 203 110 203 113 11 Poly End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.0258e-08 0.843029 -0.843029 1.0258e-08 619.253 516.221 ] concat %I 12 183 114 183 124 193 124 193 135 205 135 205 125 248 90 242 83 239 86 231 76 194 106 194 114 12 Poly End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.11572e-08 0.810855 -0.916932 9.86645e-09 659.418 526.458 ] concat %I 8 251 106 251 126 300 126 300 126 300 116 289 116 289 106 289 106 8 Poly End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.63639e-08 1.34483 -1.34483 1.63639e-08 1018.44 -282.452 ] concat %I 791 383 799 395 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 900.11 -9.08362 ] concat %I 12 776 339 776 352 798 352 803 356 808 350 807 348 812 342 808 336 803 341 789 343 782 333 776 333 12 Poly End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.13408e-08 0.952586 -0.932017 1.1591e-08 882.447 11.9457 ] concat %I 885 359 901 436 Rect End Begin %I Elli %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 1 SetP %I t [ -0.249802 0.836016 -0.836016 -0.249802 1016.8 155.898 ] concat %I 743 193 51 70 Elli End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.75 SetP %I t [ -0.289327 0.966236 -0.966236 -0.289327 1071.6 80.2259 ] concat %I 707 160 754 232 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 0.240969 0.462035 -0.606024 0.318423 615.648 549.235 ] concat %I 13 164 162 182 162 182 167 235 167 234 162 254 162 254 134 234 133 235 129 183 129 183 134 164 134 164 149 13 Poly End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 9.95725e-09 0.885621 -0.818318 1.07762e-08 598.215 291.204 ] concat %I 385 148 422 197 Rect End Begin %I Pict %I b u %I cfg Black 0 0 0 SetCFg %I cbg u %I f u %I p u %I t [ 1.05665 0 0 1.05665 213.224 -6.32959 ] concat Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg DkGray 0.501961 0.501961 0.501961 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.1286e-09 0.496511 -0.668033 6.04153e-09 312.332 583.056 ] concat %I 13 204 92 204 123 226 123 226 113 248 113 248 123 264 123 264 92 248 92 248 95 226 95 226 92 225 92 13 Poly End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg DkGray 0.501961 0.501961 0.501961 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.1286e-09 -0.496511 -0.668033 -6.04153e-09 312.332 845.214 ] concat %I 13 204 92 204 123 226 123 226 113 248 113 248 123 264 123 264 92 248 92 248 95 226 95 226 92 225 92 13 Poly End Begin %I Rect none SetB %I b n %I cfg DkGray 0.501961 0.501961 0.501961 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.1286e-09 0.496511 -0.668033 6.04153e-09 311.664 582.559 ] concat %I 258 92 274 121 Rect End End %I eop Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 532.584 731.01 ] concat %I [ (Physics) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 555.924 783.814 ] concat %I [ (Geography) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 565.058 787.873 ] concat %I [ (& Geology) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 0.737179 0.804195 -0.804195 0.737179 504.155 697.286 ] concat %I [ (Biology) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 366.956 772.951 ] concat %I [ (13th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 511.152 538.704 ] concat %I [ (Voluteer Boulevard) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 0.46602 0.986397 -0.986397 0.466021 521.533 631.373 ] concat %I [ (Middle Way) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 373.044 540.56 ] concat %I [ (16th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 363.799 672.151 ] concat %I [ (Walters) (Life) (Sciences) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 537.034 899.133 ] concat %I [ (Daughtery) (Engineering) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.60693e-08 1.32061 -1.32061 1.60693e-08 657.627 707.825 ] concat %I [ (Neyland) (Stadium) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.0536 -0.283024 0.283024 -1.0536 624.893 639.464 ] concat %I [ (Stadium Drive) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 450.384 486.441 ] concat %I [ (Library) ] Text End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.645 -1.75617 ] concat %I 4 483 431 523 431 523 391 481 389 4 Poly End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 412.866 555.498 ] concat %I [ (University) ( Center) ] Text End Begin %I BSpl %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.438 0.0836792 ] concat %I 6 753 467 837 468 841 464 846 464 843 420 841 419 6 BSpl %I 1 End Begin %I BSpl %I b 65520 2 0 0 [12 4] 17 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.921 0.182861 ] concat %I 6 788 313 830 316 840 348 841 386 843 417 843 418 6 BSpl %I 1 End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.435 0.47699 ] concat %I 887 450 839 438 Line %I 1 End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.589 0.222778 ] concat %I 9 807 460 838 460 838 405 816 404 816 394 831 394 831 379 807 378 806 379 9 Poly End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 895.838 0.0155029 ] concat %I 796 369 781 389 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 523.481 806.106 ] concat %I [ (South) (College) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 1.21714e-08 1.00029 -1.00029 1.21714e-08 439.801 711.069 ] concat %I [ (Ayres Hall) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 771.842 61.8701 ] concat %I 943 284 915 303 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 459.421 900.429 ] concat %I [ (Dabney/) (Buhler) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 767.413 57.441 ] concat %I 698 424 663 374 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-times-bold-r-*-140-* Times-Bold 14 SetF %I t [ 1.22729e-08 1.00862 -1.00862 1.22729e-08 464.596 743.589 ] concat %I [ (X) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09087 0.0128026 -0.0128027 -1.09087 503.693 612.151 ] concat %I [ (Stadium Drive) ] Text End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < cc cc 33 33 cc cc 33 33 > -1 SetP %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 855.109 167.282 ] concat %I 465 390 498 427 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 6.73658e-10 0.0553633 -0.0553633 6.73658e-10 494.25 582.008 ] concat %I 11 509 1018 237 1018 237 1114 -35 1114 -35 1018 -307 1018 -307 746 -35 746 -35 378 509 378 509 380 11 Poly End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ 1.32746e-08 1.09095 -1.09095 1.32746e-08 485.903 693.513 ] concat %I [ (Psychology) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 9.87788e-09 0.811792 -0.811792 9.87788e-09 526.522 575.313 ] concat %I [ (Parking) (Garage) ] Text End Begin %I Line %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < cc cc 33 33 cc cc 33 33 > -1 SetP %I t [ 1.07786e-08 0.885813 -0.885813 1.07786e-08 774.943 159.309 ] concat %I 481 283 491 300 Line %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 368.693 657.005 ] concat %I [ (15th Street) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -1.09095 2.65492e-08 -2.65492e-08 -1.09095 374.042 893.422 ] concat %I [ (11th Street) ] Text End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 18 476 304 494 301 501 299 527 283 545 269 569 255 590 244 611 241 648 238 672 234 686 219 705 203 741 204 767 217 776 236 774 284 776 343 773 538 18 BSpl %I 1 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 376 429 375 538 Line %I 1 End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 8 376 429 375 275 369 238 352 223 345 219 323 208 303 203 303 204 8 BSpl %I 1 End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.07785e-08 0.885813 -0.885813 1.07785e-08 786.901 202.714 ] concat %I 4 301 50 850 50 850 537 301 537 4 Poly End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 793 -5.99995 ] concat %I 5 814 52 872 66 910 82 947 107 961 127 5 BSpl %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.520591 0.958718 -0.958718 -0.520591 712.173 866.792 ] concat %I [ (Neyland Drive) ] Text End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 653.5 ] concat %I 24 459 412 422 314 224 273 224 233 224 152 305 80 345 72 386 31 418 -33 418 -122 394 -316 474 -461 652 -542 854 -566 1015 -566 1152 -501 1184 -445 1176 -203 1168 15 1031 233 789 314 547 314 432 351 428 350 24 BSpl %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 653.5 ] concat %I 916 415 915 1204 Line %I 8 End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 620.75 ] concat %I 5 486 195 402 153 394 72 402 -73 394 -73 5 BSpl %I 8 End Begin %I BSpl %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 604.375 ] concat %I 11 133 415 132 396 110 602 231 715 316 751 330 765 387 857 351 977 344 1126 358 1041 351 1190 11 BSpl %I 8 End Begin %I MLine %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 604.375 ] concat %I 3 133 410 153 -454 613 -2226 3 MLine %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 461.75 604.375 ] concat %I 663 276 137 276 Line %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 527 522.5 ] concat %I 96 105 808 79 Line %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 527 522.5 ] concat %I 95 105 -425 89 Line %I 8 End Begin %I Line %I b 65535 3 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.52099e-09 0.125 -0.125 1.52099e-09 483.5 457 ] concat %I 99 587 3984 588 Line %I 8 End Begin %I BSpl %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 6.91136e-09 0.653301 -0.567997 7.94935e-09 450.164 -70.1652 ] concat %I 15 211 347 224 320 247 286 278 265 315 254 368 252 499 251 582 255 629 257 783 266 863 302 880 318 903 384 900 434 898 545 15 BSpl %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 1.2168e-08 1 -1 1.2168e-08 288.5 532.5 ] concat %I [ (Jack Dongarra's office in Ayres Hall Room 107) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ -0.792068 1.92757e-08 -1.92757e-08 -0.792068 428.69 73.2605 ] concat %I [ (Airport/Alcoa Highway) ] Text End Begin %I Rect %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < 88 44 22 11 88 44 22 11 > -1 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 610 221 ] concat %I 258 413 267 424 Rect End Begin %I Rect %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p < 88 44 22 11 88 44 22 11 > -1 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 593 221 ] concat %I 258 413 267 424 Rect End Begin %I MLine %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 3 558 602 521 602 507 598 3 MLine %I 1 End Begin %I Line %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 326 526 326 467 Line %I 1 End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 564 368 588 385 Rect End Begin %I Rect %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 1.2168e-08 1 -1 1.2168e-08 773 26.9999 ] concat %I 564 368 588 385 Rect End Begin %I Poly %I b 65535 2 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg %I p 0.5 SetP %I t [ 8.96587e-09 0.736842 -0.736842 8.96587e-09 708.816 120.079 ] concat %I 4 616 414 631 414 631 431 616 431 4 Poly End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.855758 0.676646 -0.676646 -0.855758 381.905 599.524 ] concat %I [ (Law Builfinh) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.855758 0.676646 -0.676646 -0.855758 383.905 547.524 ] concat %I [ (Pan-Helenic Bldg.) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-80-* Courier 8 SetF %I t [ -0.855758 0.676646 -0.676646 -0.855758 384.905 572.524 ] concat %I [ (International House.) ] Text End Begin %I MLine %I b 65535 0 0 0 [] 0 SetB %I cfg Black 0 0 0 SetCFg %I cbg White 1 1 1 SetCBg none SetP %I p n %I t [ 1.2168e-08 1 -1 1.2168e-08 773 -23 ] concat %I 2 559 582 507 581 2 MLine %I 1 End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 186.965 539.756 ] concat %I [ (Ramada Inn) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-medium-r-*-100-* Courier 10 SetF %I t [ 1.13387e-08 0.931845 -0.931845 1.13387e-08 167.965 538.756 ] concat %I [ (Hilton) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ 1.2168e-08 1 -1 1.2168e-08 514.5 56.5 ] concat %I [ (Directions from the airport to Ayres Hall:) () ( Alcoa Highway North to Cumberland Avenue) () ( Cumberland Avenue east to Stadium Drive) ( \(Stadium Dr. is accross from 15th St.\)) () ( Park at Parking Garage and walk up hill) ( to largest building, Ayres Hall) ] Text End Begin %I Text %I cfg Black 0 0 0 SetCFg %I f *-courier-bold-r-*-120-* Courier-Bold 12 SetF %I t [ -0.0156231 0.999878 -0.999878 -0.0156231 651.985 47.9375 ] concat %I [ (Jack Dongarra's office phone 615-974-8295) ] Text End End %I eop showpage %%Trailer end From owner-pbwg-comm@CS.UTK.EDU Fri May 14 16:21:50 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA11143; Fri, 14 May 93 16:21:50 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA13482; Fri, 14 May 93 16:21:26 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Fri, 14 May 1993 16:21:25 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from DASHER.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA13476; Fri, 14 May 93 16:21:24 -0400 From: Jack Dongarra Received: by dasher.cs.utk.edu (5.61+IDA+UTK-930125/2.7c-UTK) id AA04396; Fri, 14 May 93 16:21:23 -0400 Date: Fri, 14 May 93 16:21:23 -0400 Message-Id: <9305142021.AA04396@dasher.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: may meeting I would like to get a rough idea how many people will attend the May 24th Parallel Benchmark Working Group meeting in Knoxville. If you are planning to attend please send me email. Thanks, Jack From owner-pbwg-comm@CS.UTK.EDU Mon May 17 04:19:22 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA14900; Mon, 17 May 93 04:19:22 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA06401; Mon, 17 May 93 04:16:33 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Mon, 17 May 1993 04:16:32 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from Mail.Think.COM by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA06388; Mon, 17 May 93 04:16:25 -0400 Received: from Godot.Think.COM by mail.think.com; Mon, 17 May 93 04:16:22 -0400 Received: by godot.think.com (4.1/Think-1.2) id AA11666; Mon, 17 May 93 04:16:21 EDT Message-Id: <9305170816.AA11666@godot.think.com> To: Jack Dongarra Cc: pbwg-comm@cs.utk.edu Subject: Re: may meeting In-Reply-To: Your message of "Fri, 14 May 93 16:21:23 EDT." <9305142021.AA04396@dasher.cs.utk.edu> Date: Mon, 17 May 93 04:16:20 EDT From: Dennis Parkinson Sorry I am unable to attend the may 24 meeting From owner-pbwg-comm@CS.UTK.EDU Tue May 18 09:30:50 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA23405; Tue, 18 May 93 09:30:50 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA07104; Tue, 18 May 93 09:30:05 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 18 May 1993 09:29:58 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from ben.uknet.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA07060; Tue, 18 May 93 09:29:47 -0400 Message-Id: <9305181329.AA07060@CS.UTK.EDU> Received: from eros.uknet.ac.uk by ben.uknet.ac.uk via UKIP with SMTP (PP) id ; Tue, 18 May 1993 14:29:31 +0100 Received: from newton.npl.co.uk by eros.uknet.ac.uk via PSS with NIFTP (PP) id <3191-0@eros.uknet.ac.uk>; Tue, 18 May 1993 14:29:28 +0100 Date: Tue, 18 May 93 14:29 GMT From: Trevor Chambers To: PBWG-COMM COMMITTEE: Whole TOPIC: Boundaries between the benchmark sets CONTENT: How big a benchmark is allowed in compact applications? ACTUAL MESSAGE: Do we need to delineate more carefully the boundaries between the benchmark sets? In particular where is the boundary between 'low level' benchmarks and 'kernels' and between `kernels' and `compact applications' What is the boundary between 'compact applications' and larger pieces of code? Trevor Chambers pp Ed Brocklehurst From owner-pbwg-comm@CS.UTK.EDU Tue May 18 13:21:52 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA26444; Tue, 18 May 93 13:21:52 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA24080; Tue, 18 May 93 13:21:18 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 18 May 1993 13:21:17 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA24067; Tue, 18 May 93 13:21:08 -0400 Via: uk.ac.southampton.ecs; Tue, 18 May 1993 17:18:11 +0100 Via: brewery.ecs.soton.ac.uk; Tue, 18 May 93 17:10:43 BST From: Tony Hey Received: from pleasuredome.ecs.soton.ac.uk by brewery.ecs.soton.ac.uk; Tue, 18 May 93 17:19:46 BST Message-Id: <11358.9305181619@pleasuredome.ecs.soton.ac.uk> Subject: What's in a name To: pbwg-comm@cs.utk.edu Date: Tue, 18 May 1993 17:19:38 +0100 (BST) X-Mailer: ELM [version 2.4 PL0] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Length: 577 Name of proposed benchmark suite Since the last meeting I have been thinking over the possibilities and doing some (random) market research. The conclusion I came to was that Pearl's suggestion: PARKBENCH - Parallel Kernels and Benchmark was really very good. Just like a SPEC mark for workstations, the idea of a PARKBENCH mark (or marks) for parallel systems seems a nice thing to say. Similarly referring to the Parkbench Suite sound OK. I think this is much better than Paraben, Interben or somesuch name. More discussion over cocktails in Knoxville? Tony Hey From owner-pbwg-comm@CS.UTK.EDU Tue May 18 13:34:13 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA26493; Tue, 18 May 93 13:34:13 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA25132; Tue, 18 May 93 13:33:48 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 18 May 1993 13:33:47 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from BERRY.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA25126; Tue, 18 May 93 13:33:46 -0400 Received: from LOCALHOST by berry.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.7c-UTK) id AA00313; Tue, 18 May 93 13:31:12 -0400 Message-Id: <9305181731.AA00313@berry.cs.utk.edu> To: Tony Hey Cc: pbwg-comm@cs.utk.edu Subject: Re: What's in a name In-Reply-To: Your message of "Tue, 18 May 1993 17:19:38 BST." <11358.9305181619@pleasuredome.ecs.soton.ac.uk> Date: Tue, 18 May 1993 13:31:11 -0400 From: "Michael W. Berry" > > > Name of proposed benchmark suite > > Since the last meeting I have been thinking over the > possibilities and doing some (random) market research. > > The conclusion I came to was that Pearl's suggestion: > > PARKBENCH - Parallel Kernels and Benchmark > > was really very good. > > Just like a SPEC mark for workstations, the idea of a > PARKBENCH mark (or marks) for parallel systems seems > a nice thing to say. Similarly referring to the > Parkbench Suite sound OK. > > I think this is much better than Paraben, Interben or > somesuch name. > > More discussion over cocktails in Knoxville? I tend to agree that PARKBENCH is very nice and would be willing to vote for it. Mike --- Michael W. Berry ___-___ o==o====== . . . . . Ayres 114 =========== ||// Department of \ \ |//__ Computer Science #_______/ berry@cs.utk.edu University of Tennessee (615) 974-3838 [OFF] Knoxville, TN 37996-1301 (615) 974-4404 [FAX] From owner-pbwg-comm@CS.UTK.EDU Wed May 26 14:04:06 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA07523; Wed, 26 May 93 14:04:06 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10932; Wed, 26 May 93 14:02:58 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 26 May 1993 14:02:57 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from BERRY.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10926; Wed, 26 May 93 14:02:55 -0400 Received: from LOCALHOST.cs.utk.edu by berry.cs.utk.edu with SMTP (5.61++/2.7c-UTK) id AA11175; Wed, 26 May 93 14:02:54 -0400 Message-Id: <9305261802.AA11175@berry.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: Minutes for Editing Date: Wed, 26 May 1993 14:02:53 -0400 From: "Michael W. Berry" Colleagues, please review the minutes below and edit where necessary. Please precede all lines that are modified/added with ">>>" so I can merge all your changes easily. Thanks, Mike B. ----------------------------------------------------------------------- Minutes of the PARKBENCH (Formerly PBWG) Workshop ------------------------------------------------- Place: Science Alliance Conference Room South College University of Tennessee Knoxville, TN Host: Jack Dongarra ORNL/Univ. of Tennessee Date: May 24, 1993 Attendees/Affiliations: - ---------------------- David Bailey, NASA Michael Berry, Univ. of Tennessee Jack Dongarra, Univ. of Tennessee / ORNL Charles Grassl Cray Research Tom Haupt, Syracuse Univ. Tony Hey, Southampton Univ. Roger Hockney, Southampton Univ. Brian LaRose, Univ. of Tennessee David Mackay, Intel SSD Joanne Martin, IBM Robert Pennington, Pittsburgh Supercomputing Center David Walker, ORNL Patrick Worley, ORNL Ramesh Natarajan, IBM, Yorktown Heights Bodo Parady, Sun Microsystems Ed Kushner, Intel SSD Agenda: May 24, 1993 - -------------------- At 8:36 am EDT, Roger Hockney gave opening remarks and welcomed all participants to the workshop. Each participant introduced him or herself by affiliation and interests. The minutes of the previous meeting (Knoxville, March 1-2) were reviewed with two major corrections made to the minutes: (1) Roger Hockney's name was missing from the Methodology subgroup list, and (2) a Compiler subgroup (with T. Haupt as leader) should have been added. Since the number of attendees was not that large (17), Roger H. proposed that there be no separate subgroup meetings during the day and all attendees agreed. Roger H. then asked the attendees to think about an alternative name for the group/benchmark suite which was first discussed at the March 1-2 meeting. The names considered include: PBWG, PARKBENCH, PARABEN, INTERBEN, SIMPLE, BIGBEN, and INTERPAR. Roger suggested that formal voting on the new name be conducted at the end of the meeting (before adjourning). The initial draft of the group's report was handed out and each chapter was then discussed in sequence. Roger H. began the discussion with Chapter 1: Methodology. David B. remarked that the the notation "Mflop/s" rather than "Mflops" (which was proposed by Roger H.) is a good standard to adopt. Although the meaning of T(p), i.e., elapsed wall-clock time on p processors, was clear, several attendees pointed out problems with the interpretation of T(1), the elapsed wall-clock time on 1 processor. Roger H. pointed out that there can be many T(1)'s, which leads to confusion in speedup comparisons. D. Bailey suggested that T(1) should not have any parallel overhead. Charles G. suggested that an efficiency measure based on Amdahl's law be used as a replacement for speedup. Ramesh N. suggested that a 2-processor baseline time be used. Tom H. pointed out that speedup is important for compiler performance measurement. Roger H. agreed that speedup may be important to report in this particular case. David B. felt that scaled speedup should be computed separately and Rameh N. indicated that "super linear" speedup is mathematically incorrect. Charles G. proposed that speedup should specifically address caching effects and memory hierarchies. This discussion of speedup ended with David B. agreeing to rewrite Section 1.4.5 (Speedup, Efficiency, and Performance per Node) of the draft and address all the concerns mentioned above. Roger H. reminded the attendees that subgroup leaders are responsible for their respective chapters of the report (which is targeted for release at Supercomputing '93). Section 1.5 (Performance Database) was the final section of Chapter 1 which was discussed. Jack D. was opposed to the idea of providing any graphical display of the on-line benchmarks provided by the PDS (Performance Database Server) extension to Xnetlib. Roger H. indicated that such graphics would make the data more attractive to users. Michael B. indicated that future PDS development would incorporate a spreadsheet-based display of the benchmark data from which graphical utilities could evolve. Tony H. indicated an interest in designing a few prototype graphical tools for displaying benchmark data obtained from PDS. Michael B. also pointed out that PDS will also provide SPEC Benchmarks in the future based on discussions with SPEC officials at a recent meeting in Huntsville, AL. Jack D. and Michael B. agreed to rewrite Section 15. and to indicate how to acquire/use PDS. Before moving on to Chapter 2 (Low-Level Benchmarks), David B. suggested that the draft include a motivation section which stresses benchmarking as a science rather than art. Parallels with other sciences could be drawn. David B. was willing to write up this for the report. Roger H. then led the discussion of Chapter 2 (Low-Level Benchmarks). He explained the difference between the two proposed timers TICK1 (clock resolution) and TICK2 (external wall-clock time). Roger also indicated that the UNIX timer "etime" is misleading in that does not report elapsed wall-clock time (reports CPU time instead) and that timer benchmarks are really necessary in order to understand the meaning of the reported times. David B. indicated that he has observed cases in which CPU time was greater than wall-clock time. Charles G. proposed that documentation should indicate that CPU time cannot be reported but also indicate potential hazards in wall-clock timing (hardware and network errors). David B. agreed to write a paragraph for the report which would address these concerns. Most attendees agreed that several runs of each benchmark should be made and Roger H. proposed that the minimum time be reported (rather than an average) for the low-level benchmarks. The consensus was unanimous on reporting the minimum time required but Bodo P. pointed out that operating systems will need to be somehow quantified for these times. Tony H. questioned whether or not optimizations should also be allowed for these particular benchmarks. A discussion of the Linpack benchmark (Section 2.1.3) and Livermore Loops (Section 2.1.4) was then initiated. Roger H. suggested that the Linpack (n=1000) benchmark be considered as a kernel benchmark. Charles G. supported the use of the Livermore Loops for measuring cache-based microprocessors and Roger H. supported their use for measuring the range of performance on a node (instability). Tony H. questioned why the group include sequential benchmarks for a parallel benchmark suite. He suggested that the report could reference the serial benchmarks (Linpack, Livermore Loops, SPEC) but should not include them in the suite. Roger H. then reviewed Section 2.1.5 which discusses the "N sub one-half" and "R sub infinity" performance measures. A routine RINF1 from the Genesis benchmarks could be used to determine these measures. Roger H. also proposed that memory-bottleneck benchmarks (POLY1, POLY2) be included (see Section 2.1.6) in the suite. Whereas vectors would fit in cache with POLY1, they would not fit in cache in POLY2. With regard to Arithmetic benchmarks (Section 2.1.7), Jack D. stressed that 64-bit arithmetic be used but Ed K. pointed out that 32-bit is commonly used in many applications (e.g., seismic codes). David B. proposed that the methodology should encourage 64-bit arithmetic but not exclude 32-bit in cases where it is explicitly required (and documented). Discussions were then curtailed for a short coffee break (10:20-10:45am). After the coffee break, discussions concerning Chapter 2 continued. Patrick W. questioned the type of communication (arbitrary or nearest- neighbor) that should be used for the COMM1, COMM2 benchmarks for measuring communication (Section 2.2). Charles G. questioned how one could measure hidden latency? Patrick W. suggested that a protocol be defined and Roger H. responded which the proposal that "nonblocking send" and "locking receive" be used. He suggested that other variations could be used in optimizing basic routines. Roger H. asked if matrix transposition really measures bisection bandwidth? David W. indicated that it does provided the matrix has only 1 data distribution. Tony H. suggested that the group think about alternative benchmarks for measuring bisection bandwidth. David W. suggested that MPI communication routines (broadcast, gather, scatter, etc.,) be used. David W. will provide information on these routines. Tony H. questioned the need for the separate communication bottleneck benchmark (POLY3, Section 2.2.4), but Roger H. maintained that it is best to have it separated from POLY1 and the COMMS benchmarks. Roger H. pointed out that the synchronization benchmark (SYNCH1) was missing from Table 2.3. Patrick W. pointed out that this particular benchmark will be extremely machine-dependent. Roger H. suggested that the basic "barrier" paradigm be used. This concluded the discussion of Chapter 2 on Low-Level Benchmarks. Roger H. then asked Tony H. to lead the discussion on Chapter 3 (Kernel Applications). Tony passed out his draft of the chapter (not included with the chapters originally handed out by Roger H.) and reviewed its contents with the attendees. For the matrix benchmarks (Section 3.2.1), Tony H. proposed that the kernel A=B*C be provided and that the group consider appropriate validation tests based on generated matrices or input datasets. It was also stressed that the matrices B and C start distributed and stay that way. Tony H. discussed the availability of a matrix diagonalization code (Intel i860) that could scale the computation per node. He will make the code available (from Dawesbury Lab) for review purposes. Jack D. proposed that routines from SCALAPACK be used for the dense LU (with pivoting) benchmark and that an iterative solver for nonsymmetric linear systems be included. Michael B., Jack D., and Patrick W. agreed to work on an appropriate sparse linear system solver or eigensolver for the suite. Jack D. suggested that a Cholesky factorization routine was not necessary as long as QR factorization was included. He also stressed that the suite use state-of-the-art algorithms for each benchmark. All attendees agreed. The discussion focused on what type of Fast Fourier Transform (FFT) benchmarks (Section 3.2.2) the suite should contain. Bodo P. suggested that they be structured like the Linpack benchmarks and questioned whether or not they should be ordered? David B. suggested that the 1-D FFT should be very large (order of 1.E+06) and need not be ordered. As an alternative, David suggested that the benchmark really be a convolution problem to be solved any way desired. Patrick W. then questioned whether or not a power of 2 should be used, and David B. responded that it should be a power or 2. Ed K. suggested that a 2-D FFT is not needed if a 3-D FFT is provided. David W. suggested that there be forward/backward FFT's which are easy to validate. Charles G. and David B. agreed to work on the FFT benchmarks. For PDE benchmarks (Section 3.2.3), there was a general agreement to drop Jacobi and Gauss-Seidel from the list of candidate algorithms. Tony H. suggested that an SOR-based routine from the Genesis benchmarks be used. David M. indicated that he could provide a Finite Element Method (FEM) code but that it might be better to consider it as a compact application rather than a kernel. Bodo P. and Tony H. proposed that such a benchmark be for a 3-D problem. There was somewhat of a consensus that there be a single problem and multiple algorithms provided. The discussion on Chapter 3 concluded with consideration of other possible kernel benchmarks (Section 3.2.4). Patrick W. questioned whether or not the Embarrassing Parallel (EP) benchmark (from NASA) should be a compact application. Roger H. suggested that there be an integer sort kernel and perhaps a Particle-In-Cell (PIC) kernel that might be commonly used in domain decomposition applications. Other suggested kernels (proposed by various attendees) included: operation counts, intrinsic operations, out-of-core solvers, check-pointing. Tony H. indicated that he could obtain an I/O benchmark from Dawesbury Lab. David B. pointed out that timing events such as loading is more appropriate for compact applications than for kernel or low-level benchmarks. Roger H. then asked David W. to lead the discussion of the final chapter of the current report (Chapter 4, Compact Applications). David B. asked if various data layouts should be allowed? David W. proposed that there should both HPF and message-passing versions of the benchmarks and questioned if time should be measured from the start to finish? Tony H. proposed that the QCD code from the Perfect Benchmarks be replaced with GAUGE (available in HPF and message-passing). Tony H. will acquire GAUGE for the Netlib database (currently listed as the "pbwg" library). David B. suggested that an N-body code be provided and Bodo P. questioned if the gravity benchmark (Section 4.2.3) is really a kernel rather than a compact application. Patrick W. indicated that he could acquire a shallow water code that is public-domain (parallelized NCAR code). David W. suggested that other molecular dynamics codes be sought since those in the Perfect Benchmarks have problem sizes that are too small. For a potential geophysics benchmark, Michael B. agreed to check on the use of the ARCO benchmark for the parallel suite. For other potential compact applications (Section 4.2.7), Bodo P. questioned the availability of DYNA2D (restricted distribution) and David M. agreed to check on the availability of a FEM code. Michael B. commented that good candidates would be those having multiple instances (HPF, message-passing, etc.,). David W. suggested that the group hold off on investigating commercial codes till the project matures. David B. suggested that a reservoir code be included. Other suggestions included: CHARMM, AMBER, GAMES, GAUSSIAN90 (all molecular dynamics codes). Patrick W. suggested that a signal processing application be added and Tony H. proposed that the applications focus under the "Grand Challenge" research areas. Charles G. asked how the performance of the compact applications would be verified. Tony H. indicated that the RAPS project typically generates lots of numbers. How many problems to run was another question raised. Bodo P. indicated that the SPEC folks use the geometric mean of several runs. Roger H. pointed out that there should be several numbers reported which can illustrate performance variation on varying numbers of processors. The discussion of Chapter 4 concluded and Roger H. asked Tom H. to briefly report on the compiler subgroup activities. Tom H. suggested that the compiler benchmarks should address how compilers handle data distribution. Related issues include the use dynamic memory, communication, and runtime libraries. He indicated that low-level compiler benchmarks (synthetic) be added and that there should be a comparison with hand-coded optimizations. Having completed formal discussions of the current report, Roger H. then called for a vote on the new name of the group. By a 10-7 margin, the attendees votes to change PBWG to PARKBENCH (PARallel Kernels and BENCHmarks). Jack D. asked if the group preferred a full-day or two half days for the next scheduled PARKBENCH meeting in Knoxville. The majority of attendees preferred the single day format, and so the next PARKBENCH meeting is scheduled to be in Knoxville on August 23. Roger H. asked that the minutes be posted (to comp.parallel Internet newsgroup and pbwg-com@cs.utk.edu). Joanne M. indicated that she would have information on the Birds-of-a-Feather (BOF) session for PARKBENCH at Supercomputing '93 at the August meeting. Michael B. briefly reviewed the status of the SPEC/Perfect merger and passed out minutes of that meeting (Huntsville, May 1-13). Roger H. then adjourned the official (third) meeting of the PARKBENCH group at 2:50pm EDT. A demo of the PDS tool supported by UT/ORNL was given by Brian L. to a few of the attendees till approximately 3:15pm EDT. From owner-pbwg-comm@CS.UTK.EDU Fri May 28 21:01:01 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA22536; Fri, 28 May 93 21:01:01 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA20147; Fri, 28 May 93 21:00:58 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Fri, 28 May 1993 21:00:57 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from BERRY.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA20141; Fri, 28 May 93 21:00:55 -0400 Received: from LOCALHOST.cs.utk.edu by berry.cs.utk.edu with SMTP (5.61++/2.7c-UTK) id AA17854; Fri, 28 May 93 21:00:54 -0400 Message-Id: <9305290100.AA17854@berry.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: Revised Minutes of last Meeting Date: Fri, 28 May 1993 21:00:54 -0400 From: "Michael W. Berry" Minutes of the PARKBENCH (Formerly PBWG) Workshop ------------------------------------------------- (PBWG= Parallel Benchmark Working Group) Place: Science Alliance Conference Room South College University of Tennessee Knoxville, TN Host: Jack Dongarra ORNL/Univ. of Tennessee Date: May 24, 1993 Attendees/Affiliations: ---------------------- David Bailey, NASA Michael Berry, Univ. of Tennessee Jack Dongarra, Univ. of Tennessee / ORNL Charles Grassl Cray Research Tom Haupt, Syracuse Univ. Tony Hey, Southampton Univ. Roger Hockney, Southampton Univ. Ed Kushner, Intel SSD Brian LaRose, Univ. of Tennessee David Mackay, Intel SSD Joanne Martin, IBM Ramesh Natarajan, IBM, Yorktown Heights Robert Pennington, Pittsburgh Supercomputing Center Bodo Parady, Sun Microsystems David Walker, ORNL Patrick Worley, ORNL Agenda: May 24, 1993 -------------------- At 8:36 am EDT, Roger Hockney gave opening remarks and welcomed all participants to the workshop. Each participant introduced him or herself by affiliation and interests. The minutes of the previous meeting (Knoxville, March 1-2) were reviewed with two major corrections made to the minutes: (1) Roger Hockney's name was missing from the Methodology subgroup list, and (2) a Compiler subgroup (with T. Haupt as leader) should have been added. Since the number of attendees was not that large (17), Roger H. proposed that there be no separate subgroup meetings during the day and all attendees agreed. Roger H. then asked the attendees to think about an alternative name for the group/benchmark suite which was first discussed at the March 1-2 meeting. The names considered include: PBWG, PARKBENCH, PARABEN, INTERBEN, SIMPLE, BIGBEN, and INTERPAR. Roger suggested that formal voting on the new name be conducted at the end of the meeting (before adjourning). The initial draft of the group's report was handed out and each chapter was then discussed in sequence. Roger H. began the discussion with Chapter 1: Methodology. David B. remarked that the the notation "Mflop/s" rather than "Mflops" (which was proposed by Roger H.) is a good standard to adopt. Although the meaning of T(p), i.e., elapsed wall-clock time on p processors, was clear, several attendees pointed out problems with the interpretation of T(1), the elapsed wall-clock time on 1 processor. Roger H. pointed out that there can be many T(1)'s, which leads to confusion in speedup comparisons. D. Bailey suggested that T(1) should not have any parallel overhead. Charles G. suggested that an efficiency measure based on Amdahl's law be used as a replacement for speedup. Ramesh N. suggested that a 2-processor baseline time be used. Tom H. pointed out that speedup is important for compiler performance measurement. Roger H. agreed that speedup may be important to report in this particular case. Ramesh N. felt that scaled speedup should be computed separately and that the widespread connotation of the term "superlinear" speedup is mathematically incorrect. Charles G. proposed that speedup should specifically address caching effects and memory hierarchies. This discussion of speedup ended with David B. agreeing to rewrite Section 1.4.5 (Speedup, Efficiency, and Performance per Node) of the draft and address all the concerns mentioned above. Roger H. reminded the attendees that subgroup leaders are responsible for their respective chapters of the report (which is targeted for release at Supercomputing '93). Section 1.5 (Performance Database) was the final section of Chapter 1 which was discussed. Jack D. was opposed to the idea of providing any graphical display of the on-line benchmarks provided by the PDS (Performance Database Server) extension to Xnetlib. Roger H. indicated that such graphics would make the data more attractive to users. Michael B. indicated that future PDS development would incorporate a spreadsheet-based display of the benchmark data from which graphical utilities could evolve. Tony H. indicated an interest in designing a few prototype graphical tools for displaying benchmark data obtained from PDS. Michael B. also pointed out that PDS will also provide SPEC Benchmarks in the future based on discussions with SPEC officials at a recent meeting in Huntsville, AL. Jack D. and Michael B. agreed to rewrite Section 15. and to indicate how to acquire/use PDS. Before moving on to Chapter 2 (Low-Level Benchmarks), David B. suggested that the draft include a motivation section which stresses benchmarking as a science rather than art. Parallels with other sciences could be drawn. David B. was willing to write up this for the report. Roger H. then led the discussion of Chapter 2 (Low-Level Benchmarks). He explained the difference between the two proposed timers TICK1 (clock resolution) and TICK2 (external wall-clock time). Roger also indicated that the UNIX timer "etime" is misleading in that does not report elapsed wall-clock time (reports CPU time instead) and that timer benchmarks are really necessary in order to understand the meaning of the reported times. David B. indicated that he has observed cases in which CPU time was greater than wall-clock time. Charles G. proposed that documentation should indicate that CPU time cannot be reported but also indicate potential hazards in wall-clock timing (hardware and network errors). David B. agreed to write a paragraph for the report which would address these concerns. Most attendees agreed that several runs of each benchmark should be made and Roger H. proposed that the minimum time be reported (rather than an average) for the low-level benchmarks. The consensus was unanimous on reporting the minimum time required but Bodo P. pointed out that operating systems will need to be somehow quantified for these times. Tony H. questioned whether or not optimizations should also be allowed for these particular benchmarks. A discussion of the Linpack benchmark (Section 2.1.3) and Livermore Loops (Section 2.1.4) was then initiated. Roger H. suggested that the Linpack (n=1000) benchmark be considered as a kernel benchmark. Charles G. did not support the use of the Livermore Loops for measuring cache-based microprocessors, while Roger H. supported their use for measuring the range of performance on a node (instability). Tony H. questioned why the group include sequential benchmarks for a parallel benchmark suite. He suggested that the report could reference the serial benchmarks (Linpack, Livermore Loops, SPEC) but should not include them in the suite. Roger H. then reviewed Section 2.1.5 which discusses the "N sub one-half" and "R sub infinity" performance measures. A routine RINF1 from the Genesis benchmarks could be used to determine these measures. Roger H. also proposed that memory-bottleneck benchmarks (POLY1, POLY2) be included (see Section 2.1.6) in the suite. Whereas vectors would fit in cache with POLY1, they would not fit in cache in POLY2. With regard to Arithmetic benchmarks (Section 2.1.7), Jack D. stressed that 64-bit arithmetic be used but Ed K. pointed out that 32-bit is commonly used in many applications (e.g., seismic codes). David B. proposed that the methodology should encourage 64-bit arithmetic but not exclude 32-bit in cases where it is explicitly required (and documented). Discussions were then curtailed for a short coffee break (10:20-10:45am). After the coffee break, discussions concerning Chapter 2 continued. Patrick W. questioned the type of communication (arbitrary or nearest- neighbor) that should be used for the COMM1, COMM2 benchmarks for measuring communication (Section 2.2). Charles G. enquired as to if one could measure hidden latency? Patrick W. suggested that a protocol be defined and Roger H. responded which the proposal that "nonblocking send" and "locking receive" be used. He suggested that other variations could be used in optimizing basic routines. Roger H. asked if matrix transposition really measures bisection bandwidth? David W. indicated that it does provided the matrix has only 1 data distribution. Tony H. suggested that the group think about alternative benchmarks for measuring bisection bandwidth. David W. suggested that MPI communication routines (broadcast, gather, scatter, etc.,) be used. David W. will provide information on these routines. Tony H. questioned the need for the separate communication bottleneck benchmark (POLY3, Section 2.2.4), but Roger H. maintained that it is best to have it separated from POLY1 and the COMMS benchmarks. Roger H. pointed out that the synchronization benchmark (SYNCH1) was missing from Table 2.3. Patrick W. pointed out that this particular benchmark will be extremely machine-dependent. Roger H. suggested that the basic "barrier" paradigm be used. This concluded the discussion of Chapter 2 on Low-Level Benchmarks. Roger H. then asked Tony H. to lead the discussion on Chapter 3 (Kernel Applications). Tony passed out his draft of the chapter (not included with the chapters originally handed out by Roger H.) and reviewed its contents with the attendees. For the matrix benchmarks (Section 3.2.1), Tony H. proposed that the kernel A=B*C be provided and that the group consider appropriate validation tests based on generated matrices or input datasets. It was also stressed that the matrices B and C start distributed and stay that way. Tony H. discussed the availability of a matrix diagonalization code (Intel i860) that could scale the computation per node. He will make the code available (from Dawesbury Lab) for review purposes. Jack D. proposed that routines from SCALAPACK be used for the dense LU (with pivoting) benchmark and that an iterative solver for nonsymmetric linear systems be included. Michael B., Jack D., and Patrick W. agreed to work on an appropriate sparse linear system solver or eigensolver for the suite. Jack D. suggested that a Cholesky factorization routine was not necessary as long as QR factorization was included. He also stressed that the suite use state-of-the-art algorithms for each benchmark. All attendees agreed. The discussion focused on what type of Fast Fourier Transform (FFT) benchmarks (Section 3.2.2) the suite should contain. Bodo P. suggested that they be structured like the Linpack benchmarks and questioned whether or not they should be ordered? David B. suggested that the 1-D FFT should be very large (order of 1.E+06) and need not be ordered. As an alternative, David suggested that the benchmark really be a convolution problem to be solved any way desired. Patrick W. then questioned whether or not a power of 2 should be used, and David B. responded that it should be a power or 2. Ed K. suggested that a 2-D FFT is not needed if a 3-D FFT is provided. David W. suggested that there be forward/backward FFT's which are easy to validate. Charles G. and David B. agreed to work on the FFT benchmarks. For PDE benchmarks (Section 3.2.3), there was a general agreement to drop Jacobi and Gauss-Seidel from the list of candidate algorithms. Tony H. suggested that an SOR-based routine from the Genesis benchmarks be used. David M. indicated that he could provide a Finite Element Method (FEM) code but that it might be better to consider it as a compact application rather than a kernel. Bodo P. and Tony H. proposed that such a benchmark be for a 3-D problem. There was somewhat of a consensus that there be a single problem and multiple algorithms provided. The discussion on Chapter 3 concluded with consideration of other possible kernel benchmarks (Section 3.2.4). Patrick W. questioned whether or not the Embarrassing Parallel (EP) benchmark (from NASA) should be a compact application. Roger H. suggested that there be an integer sort kernel and perhaps a Particle-In-Cell (PIC) kernel that might be commonly used in domain decomposition applications. Other suggested kernels (proposed by various attendees) included: operation counts, intrinsic operations, out-of-core solvers, check-pointing. Tony H. indicated that he could obtain an I/O benchmark from Dawesbury Lab. David B. pointed out that timing events such as loading is more appropriate for compact applications than for kernel or low-level benchmarks. Roger H. then asked David W. to lead the discussion of the final chapter of the current report (Chapter 4, Compact Applications). David B. asked if various data layouts should be allowed? David W. proposed that there should both HPF and message-passing versions of the benchmarks and questioned if time should be measured from the start to finish? Tony H. proposed that the QCD code from the Perfect Benchmarks be replaced with GAUGE (available in HPF and message-passing). Tony H. will acquire GAUGE for the Netlib database (currently listed as the "pbwg" library). David B. suggested that an N-body code be provided and Bodo P. questioned if the gravity benchmark (Section 4.2.3) is really a kernel rather than a compact application. Patrick W. indicated that he could acquire a shallow water code that is public-domain (parallelized NCAR code). David W. suggested that other molecular dynamics codes be sought since those in the Perfect Benchmarks have problem sizes that are too small. For a potential geophysics benchmark, Michael B. agreed to check on the use of the ARCO benchmark for the parallel suite. For other potential compact applications (Section 4.2.7), Bodo P. objected to the use of DYNA3D (originally suggested by Joanne M.), and David M. agreed to check on the availability of a FEM code. Michael B. commented that good candidates would be those having multiple instances (HPF, message-passing, etc.,). David W. suggested that the group hold off on investigating commercial codes till the project matures. David B. suggested that a reservoir code be included. Other suggestions included: CHARMM, AMBER, GAMES, GAUSSIAN90 (all molecular dynamics codes). Patrick W. suggested that a signal processing application be added and Tony H. proposed that the applications focus under the "Grand Challenge" research areas. Charles G. asked how the performance of the compact applications would be verified. Tony H. indicated that the RAPS project typically generates lots of numbers. How many problems to run was another question raised. Bodo P. indicated that the SPEC folks use the geometric mean of several runs. Roger H. pointed out that there should be several numbers reported which can illustrate performance variation on varying numbers of processors. The discussion of Chapter 4 concluded and Roger H. asked Tom H. to briefly report on the compiler subgroup activities. Tom H. suggested that the compiler benchmarks should address how compilers handle data distribution. Related issues include the use dynamic memory, communication, and runtime libraries. He indicated that low-level compiler benchmarks (synthetic) be added and that there should be a comparison with hand-coded optimizations. Having completed formal discussions of the current report, Roger H. then called for a vote on the new name of the group. By a 10-7 margin, the attendees voted to change PBWG to PARKBENCH (PARallel Kernels and BENCHmarks). Jack D. asked if the group preferred a full-day or two half days for the next scheduled PARKBENCH meeting in Knoxville. The majority of attendees preferred the single day format, and so the next PARKBENCH meeting is scheduled to be in Knoxville on August 23. Roger H. asked that the minutes be posted (to comp.parallel Internet newsgroup and pbwg-com@cs.utk.edu). Joanne M. indicated that she would have information on the Birds-of-a-Feather (BOF) session for PARKBENCH at Supercomputing '93 at the August meeting. Michael B. briefly reviewed the status of the SPEC/Perfect merger and passed out minutes of that meeting (Huntsville, May 1-13). Roger H. then adjourned the official (third) meeting of the PARKBENCH group at 2:50pm EDT. A demo of the PDS tool supported by UT/ORNL was given by Brian L. to a few of the attendees till approximately 3:15pm EDT. End of Minutes for May 24, 1993 (M. Berry) --- From owner-pbwg-comm@CS.UTK.EDU Tue Jun 1 17:58:44 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA05508; Tue, 1 Jun 93 17:58:44 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA26587; Tue, 1 Jun 93 17:58:22 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Tue, 1 Jun 1993 17:58:21 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from BERRY.CS.UTK.EDU by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA26581; Tue, 1 Jun 93 17:58:20 -0400 Received: from LOCALHOST.cs.utk.edu by berry.cs.utk.edu with SMTP (5.61++/2.7c-UTK) id AA03189; Tue, 1 Jun 93 17:58:19 -0400 Message-Id: <9306012158.AA03189@berry.cs.utk.edu> To: pbwg-comm@cs.utk.edu Subject: Revised Minutes Date: Tue, 01 Jun 1993 17:58:18 -0400 From: "Michael W. Berry" This should be the last revision - please let me know if I have missed any other changes. -Mike -----------------------------START MINUTES---------------------------- Minutes of the 3rd PARKBENCH (Formerly PBWG) Workshop ----------------------------------------------------- (PBWG= Parallel Benchmark Working Group) Place: Science Alliance Conference Room South College University of Tennessee Knoxville, TN Host: Jack Dongarra ORNL/Univ. of Tennessee Date: May 24, 1993 Attendees/Affiliations: ---------------------- David Bailey, NASA Michael Berry, Univ. of Tennessee Jack Dongarra, Univ. of Tennessee / ORNL Charles Grassl Cray Research Tom Haupt, Syracuse Univ. Tony Hey, Southampton Univ. Roger Hockney, Southampton Univ. Ed Kushner, Intel SSD Brian LaRose, Univ. of Tennessee David Mackay, Intel SSD Joanne Martin, IBM Ramesh Natarajan, IBM, Yorktown Heights Robert Pennington, Pittsburgh Supercomputing Center Bodo Parady, Sun Microsystems David Walker, ORNL Patrick Worley, ORNL Agenda: May 24, 1993 -------------------- At 8:36 am EDT, Roger Hockney gave opening remarks and welcomed all participants to the workshop. Each participant introduced him or herself by affiliation and interests. The minutes of the previous meeting (Knoxville, March 1-2) were reviewed with two major corrections made to the minutes: (1) Roger Hockney's name was missing from the Methodology subgroup list, and (2) a Compiler subgroup (with T. Haupt as leader) should have been added. Since the number of attendees was not that large (17), Roger H. proposed that there be no separate subgroup meetings during the day and all attendees agreed. Roger H. then asked the attendees to think about an alternative name for the group/benchmark suite which was first discussed at the March 1-2 meeting. The names considered include: PBWG, PARKBENCH, PARABEN, INTERBEN, SIMPLE, BIGBEN, and INTERPAR. Roger suggested that formal voting on the new name be conducted at the end of the meeting (before adjourning). The initial draft of the group's report was handed out and each chapter was then discussed in sequence. Roger H. began the discussion with Chapter 1: Methodology. David B. remarked that the the notation "Mflop/s" rather than "Mflops" (which was proposed by Roger H.) is a good standard to adopt. Although the meaning of T(p), i.e., elapsed wall-clock time on p processors, was clear, several attendees pointed out problems with the interpretation of T(1), the elapsed wall-clock time on 1 processor. Roger H. pointed out that there can be many T(1)'s, which leads to confusion in speedup comparisons. D. Bailey suggested that T(1) should not have any parallel overhead. Charles G. suggested that an efficiency measure based on Amdahl's law be used as a replacement for speedup. Ramesh N. suggested that a 2-processor baseline time be used. Tom H. pointed out that speedup is important for compiler performance measurement. Roger H. agreed that speedup may be important to report in this particular case. Ramesh N. felt that scaled speedup should be computed separately and that the widespread connotation of the term "superlinear" speedup is mathematically incorrect. Charles G. proposed that speedup should specifically address caching effects and memory hierarchies. This discussion of speedup ended with David B. agreeing to rewrite Section 1.4.5 (Speedup, Efficiency, and Performance per Node) of the draft and address all the concerns mentioned above. Roger H. reminded the attendees that subgroup leaders are responsible for their respective chapters of the report (which is targeted for release at Supercomputing '93). Section 1.5 (Performance Database) was the final section of Chapter 1 which was discussed. Jack D. was opposed to the idea of providing any graphical display of the on-line benchmarks provided by the PDS (Performance Database Server) extension to Xnetlib. Roger H. indicated that such graphics would make the data more attractive to users. Michael B. indicated that future PDS development would incorporate a spreadsheet-based display of the benchmark data from which graphical utilities could evolve. Tony H. indicated an interest in designing a few prototype graphical tools for displaying benchmark data obtained from PDS. Michael B. also pointed out that PDS will also provide SPEC Benchmarks in the future based on discussions with SPEC officials at a recent meeting in Huntsville, AL. Jack D. and Michael B. agreed to rewrite Section 15. and to indicate how to acquire/use PDS. Before moving on to Chapter 2 (Low-Level Benchmarks), David B. suggested that the draft include a motivation section which stresses benchmarking as a science rather than art. Parallels with other sciences could be drawn. David B. was willing to write up this for the report. Roger H. then led the discussion of Chapter 2 (Low-Level Benchmarks). He explained the difference between the two proposed timers TICK1 (clock resolution) and TICK2 (external wall-clock time). Roger also indicated that the UNIX timer "etime" is misleading in that does not report elapsed wall-clock time (reports CPU time instead) and that timer benchmarks are really necessary in order to understand the meaning of the reported times. David B. indicated that he has observed cases in which CPU time was greater than wall-clock time. Charles G. proposed that documentation should indicate that CPU time cannot be reported but also indicate potential hazards in wall-clock timing (hardware and network errors). David B. agreed to write a paragraph for the report which would address these concerns. Most attendees agreed that several runs of each benchmark should be made and Roger H. proposed that the minimum time be reported (rather than an average) for the low-level benchmarks. The consensus was unanimous on reporting the minimum time required but Bodo P. pointed out that operating systems will need to be somehow quantified for these times. Tony H. questioned whether or not optimizations should also be allowed for these particular benchmarks. A discussion of the Linpack benchmark (Section 2.1.3) and Livermore Loops (Section 2.1.4) was then initiated. Roger H. suggested that the Linpack (n=1000) benchmark be considered as a kernel benchmark. Charles G. did not support the use of the Livermore Loops for measuring cache-based microprocessors, while Roger H. supported their use for measuring the range of performance on a node (instability). Tony H. questioned why the group include sequential benchmarks for a parallel benchmark suite. He suggested that the report could reference the serial benchmarks (Linpack, Livermore Loops, SPEC) but should not include them in the suite. Roger H. then reviewed Section 2.1.5 which discusses the "N sub one-half" and "R sub infinity" performance measures. A routine RINF1 from the Genesis benchmarks could be used to determine these measures. Roger H. also proposed that memory-bottleneck benchmarks (POLY1, POLY2) be included (see Section 2.1.6) in the suite. Whereas vectors would fit in cache with POLY1, they would not fit in cache in POLY2. With regard to Arithmetic benchmarks (Section 2.1.7), Jack D. stressed that 64-bit arithmetic be used but Ed K. pointed out that 32-bit is commonly used in many applications (e.g., seismic codes). David B. proposed that the methodology should encourage 64-bit arithmetic but not exclude 32-bit in cases where it is explicitly required (and documented). Discussions were then curtailed for a short coffee break (10:20-10:45am). After the coffee break, discussions concerning Chapter 2 continued. Patrick W. questioned the type of communication (arbitrary or nearest- neighbor) that should be used for the COMM1, COMM2 benchmarks for measuring communication (Section 2.2). Charles G. enquired as to if one could measure hidden latency? Patrick W. suggested that a protocol be defined and Roger H. responded with the proposal that "nonblocking send" and "blocking receive" be used. He suggested that other variations could be used in optimizing basic routines. Roger H. asked if matrix transposition really measures bisection bandwidth? David W. indicated that it does provided the matrix has only 1 data distribution. Tony H. suggested that the group think about alternative benchmarks for measuring bisection bandwidth. David W. suggested that MPI communication routines (broadcast, gather, scatter, etc.,) be used. David W. will provide information on these routines. Tony H. questioned the need for the separate communication bottleneck benchmark (POLY3, Section 2.2.4), but Roger H. maintained that it is best to have it separated from POLY1 and the COMMS benchmarks. Roger H. pointed out that the synchronization benchmark (SYNCH1) was missing from Table 2.3. Patrick W. pointed out that this particular benchmark will be extremely machine-dependent. Roger H. suggested that the basic "barrier" paradigm be used. This concluded the discussion of Chapter 2 on Low-Level Benchmarks. Roger H. then asked Tony H. to lead the discussion on Chapter 3 (Kernel Applications). Tony passed out his draft of the chapter (not included with the chapters originally handed out by Roger H.) and reviewed its contents with the attendees. For the matrix benchmarks (Section 3.2.1), Tony H. proposed that the kernel A=B*C be provided and that the group consider appropriate validation tests based on generated matrices or input datasets. It was also stressed that the matrices B and C start distributed and stay that way. Tony H. discussed the availability of a matrix diagonalization code (Intel i860) that could scale the computation per node. He will make the code available (from Daresbury Lab) for review purposes. Jack D. proposed that routines from SCALAPACK be used for the dense LU (with pivoting) benchmark and that an iterative solver for nonsymmetric linear systems be included. Michael B., Jack D., and Patrick W. agreed to work on an appropriate sparse linear system solver or eigensolver for the suite. Jack D. suggested that a Cholesky factorization routine was not necessary as long as QR factorization was included. He also stressed that the suite use state-of-the-art algorithms for each benchmark. All attendees agreed. The discussion focused on what type of Fast Fourier Transform (FFT) benchmarks (Section 3.2.2) the suite should contain. Bodo P. suggested that they be structured like the Linpack benchmarks and questioned whether or not they should be ordered? David B. suggested that the 1-D FFT should be very large (order of 1.E+06) and need not be ordered. As an alternative, David suggested that the benchmark really be a convolution problem to be solved any way desired. Patrick W. then questioned whether or not a power of 2 should be used, and David B. responded that it should be a power or 2. Ed K. suggested that a 2-D FFT is not needed if a 3-D FFT is provided. David W. suggested that there be forward/backward FFT's which are easy to validate. Charles G. and David B. agreed to work on the FFT benchmarks. For PDE benchmarks (Section 3.2.3), there was a general agreement to drop Jacobi and Gauss-Seidel from the list of candidate algorithms. Tony H. suggested that an SOR-based routine from the Genesis benchmarks be used. David M. indicated that he could provide a Finite Element Method (FEM) code but that it might be better to consider it as a compact application rather than a kernel. Bodo P. and Tony H. proposed that such a benchmark be for a 3-D problem. There was somewhat of a consensus that there be a single problem and multiple algorithms provided. The discussion on Chapter 3 concluded with consideration of other possible kernel benchmarks (Section 3.2.4). Patrick W. questioned whether or not the Embarrassing Parallel (EP) benchmark (from NASA) should be a compact application. Roger H. suggested that there be an integer sort kernel and perhaps a Particle-In-Cell (PIC) kernel that might be commonly used in domain decomposition applications. Other suggested kernels (proposed by various attendees) included: operation counts, intrinsic operations, out-of-core solvers, check-pointing. Tony H. indicated that he could obtain an I/O benchmark from Daresbury Lab. David B. pointed out that timing events such as loading is more appropriate for compact applications than for kernel or low-level benchmarks. Roger H. then asked David W. to lead the discussion of the final chapter of the current report (Chapter 4, Compact Applications). David B. asked if various data layouts should be allowed? David W. proposed that there should both HPF and message-passing versions of the benchmarks and questioned if time should be measured from the start to finish? Tony H. proposed that the QCD code from the Perfect Benchmarks be replaced with GAUGE (available in HPF and message-passing). Tony H. will acquire GAUGE for the Netlib database (currently listed as the "pbwg" library). David B. suggested that an N-body code be provided and Bodo P. questioned if the gravity benchmark (Section 4.2.3) is really a kernel rather than a compact application. Patrick W. indicated that he could acquire a shallow water code that is public-domain (parallelized NCAR code). David W. suggested that other molecular dynamics codes be sought since those in the Perfect Benchmarks have problem sizes that are too small. For a potential geophysics benchmark, Michael B. agreed to check on the use of the ARCO benchmark for the parallel suite. For other potential compact applications (Section 4.2.7), Bodo P. objected to the use of DYNA3D (originally suggested by Joanne M.), and David M. agreed to check on the availability of a FEM code. Michael B. commented that good candidates would be those having multiple instances (HPF, message-passing, etc.,). David W. suggested that the group hold off on investigating commercial codes till the project matures. David B. suggested that a reservoir code be included. Other suggestions included: CHARMM, AMBER, GAMES, GAUSSIAN90 (all molecular dynamics codes). Patrick W. suggested that a signal processing application be added and Tony H. proposed that the applications focus under the "Grand Challenge" research areas. Charles G. asked how the performance of the compact applications would be verified. Tony H. indicated that the RAPS project typically generates lots of numbers. How many problems to run was another question raised. Bodo P. indicated that the SPEC folks use the geometric mean of several runs. Roger H. pointed out that there should be sufficient numbers reported to be able to show the performance variation with the numbers of processors. It is particularly important to detect Amdahl saturation, and any peak in performance which is followed by a steady decrease. Between 5 and 10 points for each problem size, roughly equally spaced logarithmically, would usually be necessary to do this (e.g. 1, 2, 4, 8, 16, 32, 64 and 128, on a 128-node system). Such detailed measurements have been made without difficulty on the LPM1 benchmark, and give a clear picture of performance variation with both problem size and number of processors. The discussion of Chapter 4 concluded and Roger H. asked Tom H. to briefly report on the compiler subgroup activities. Tom H. suggested that the compiler benchmarks should address how compilers handle data distribution. Related issues include the use dynamic memory, communication, and runtime libraries. He indicated that low-level compiler benchmarks (synthetic) be added and that there should be a comparison with hand-coded optimizations. Having completed formal discussions of the current report, Roger H. then called for a vote on the new name of the group. By a 10-7 margin, the attendees voted to change PBWG to PARKBENCH (PARallel Kernels and BENCHmarks). Jack D. asked if the group preferred a full-day or two half days for the next scheduled PARKBENCH meeting in Knoxville. The majority of attendees preferred the single day format, and so the next PARKBENCH meeting is scheduled to be in Knoxville on August 23. Roger H. asked that the minutes be posted (to comp.parallel Internet newsgroup and pbwg-com@cs.utk.edu). Joanne M. indicated that she would have information on the Birds-of-a-Feather (BOF) session for PARKBENCH at Supercomputing '93 at the August meeting. Michael B. briefly reviewed the status of the SPEC/Perfect merger and passed out minutes of that meeting (Huntsville, May 1-13). Roger H. then adjourned the official (third) meeting of the PARKBENCH group at 2:50pm EDT. A demo of the PDS tool supported by UT/ORNL was given by Brian L. to a few of the attendees till approximately 3:15pm EDT. End of Minutes for May 24, 1993 (M. Berry) From owner-pbwg-comm@CS.UTK.EDU Wed Jun 2 14:21:29 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA11051; Wed, 2 Jun 93 14:21:29 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA15612; Wed, 2 Jun 93 14:21:24 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 2 Jun 1993 14:21:19 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sun2.nsfnet-relay.ac.uk by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA15592; Wed, 2 Jun 93 14:21:13 -0400 Via: uk.ac.southampton.ecs; Wed, 2 Jun 1993 18:00:21 +0100 From: R.Hockney@parallel-applications-centre.southampton.ac.uk Via: calvados.pac.soton.ac.uk (plonk); Wed, 2 Jun 93 17:52:29 BST Date: Wed, 2 Jun 93 16:59:55 GMT Message-Id: <18087.9306021659@calvados.pac.soton.ac.uk> To: pbwg-comm@cs.utk.edu Subject: Revised SPEEDUP section Because the definition of Speedup is of such general concern to all members, not just those in the Methodology subcommittee, I am sending this proposed ammendment to all committee members: *********************** \subsection{Speedup and Efficiency} Speedup is a popular metric that has been used for many years to compare the performance of parallel computers. However its definition is open to ambiguity and misuse because it always begs the question "Speedup over what?"; a question that is often not clearly answered in publications using this metric. Whilst preferring the use of absolute measures of performance, such as Benchmark Performance defined earlier, the PARKBENCH committee accepted that speedup would probably continue to be used, and that the best policy was to sharpen-up its definition. Speedup is universally defined as \begin{equation} frac{T_1}{T_p} \end{equation} where $T_p$ is the p-processor time to perform some benchmark, and $T_1$ is called the one-processor time. There is no doubt about the meaning of $T_p$, because this is measured time $T(N;p)$ to perform the benchmark. There is, however, usually considerable discussion over the meaning of $T_1$, whether it is the time for the parallel code running on one-processor which probably contains unnecessary parallel overheads, or whether it is the best serial code (probably quite a different algorithm) running on one processor. The latter choice sounds much more realistic, but would require a program of research to determine what was the best serial algorithm, and the rescaling of all previously computed Speedup values every time a better serial algorithm was discovered. An additional problem is that even if we decide what $T_1$ should be, there may not be enough memory on a single node to store the whole data for a large problem suitable for using a large MPP. It may not therefore be possible to measure $T_1$ on an MPP, however we define it. The purpose of benchmarking is to compare the performance of different computers, on the basis that the best performance corresponds to the least wall-clock execution time. In order to use Speedup for this purpose, it does not matter how $T_1$ is defined, or what its value is. It only matters that the same value of $T_1$ is used to calculate all Speedup values used in the comparison. Looked at in this way, $T_1$ is just a single reference time which is defined for each benchmark, and to which all parallel execution times are compared. The answer to the question "over what?" is then "over $T_1$", and it is clear then why the same $T_1$ must be used for all comparisons of different computers on the same benchmark. If we do not use the same $T_1$ for all comparisons, then we are using different units to measure the performance on the different computers. This makes as much sense as comparing the numerical value of the maximum speeds of three cars, when one is measured in m.p.h, the second in feet per second and the third in m/s. The SPEC committee uses the above proceedure in the definition of their SPEC ratio which is defined as the Speedup over a reference time obtained by running the defining serial code on a VAX11/780. The problem is that periodically it seems necessary to update these reference times to a currently available computer, or to keep a VAX11/780 going in a special museum (I suppose it would be the Smithsonian or NPL) in a similar way as the standard yard or metre are carefully maintained. Nevertheless there is no doubt that keeping a constant value of $T_1$, however it is defined, for each benchmark is the only way of making Speedup an acceptable metric for measuring and comparing computer performance. Defining $T_1$ as a reference time unrelated to the parallel computer being benchmarked unfortunately has the conseqence that certain properties that many people regard as essential to the idea of Speedup are lost: \begin{enumerate} \item It is no longer necessarilly true that the Speedup of the parallel code on one processor is unity. It may be, but only by chance. \item It is no longer true that the maximum Speedup using $p$-processors is $p$. \item Because of the last item, Efficiency=Speedup/$p$ is no longer a meaningful measure of processor utilisation. \end{enumerate} Thus it appears that if we sharpen-up the definition of Speedup to make it an acceptable metric for comparing the performance of different computers, we have to throw way the main properties which have made the concept of Speedup useful in the past. There is a choice: keep Speedup with its traditional properties, and accept that it has no place as a metric for comparing computer performance (i.e. in benchmarking), or define Speedup in a way that can be used in benchmarking, and lose the traditional properties. There is no middle way, or possible compromise. If we use $T_1$ as the time for the defining serial code on a very fast single processor (currently, say, a CRAY C90), then I am sure that manufacturers would be reluctant to having to quote the Speedup of their MPP with hundreds of processors in the above way. If the Speedup of the 100 processor MPP over a single node of the MPP is a respectable 80, say, it is likely that the Speedup over $T_1$ would be reduced to about 10 or less, because the fast single processor is likely to be at least ten times faster than the workstation chips used in MPPs. [For all the above reasons I, personally RWH, do not believe that Speedup can be saved as a useful metric for comparing computer performance, and that it should only be kept as a convenient metric to use when optimising code on a particular multiprocessor computer in isolation. However if the committee wishes to allow its use as a metric the following rules should apply:] The use of absolute measures of computer performance such as Temporal (tstep/s) or Benchmark performance (Mflop/s based on a given nominal flop-count) avoid the above problems of definition. However, if Speedup is used as a metric for comparing computer performance on benchmarks, then the PARKBENCH committee requires that: \begin{enumerate} \item The value of $T_1$ in seconds that was used in the speedup calculation must be quoted along with the value of Speedup \item The same value of $T_1$ must be used when calculating all Speedup values for a particular benchmark. \item The benchmark writer must provide, as part of the definition of the benchmark, the value of $T_1$ in seconds that is to be used. \item Only if the above rules are obeyed will the benchmark results be accepted as unambiguous and entered into the PDS database. \end{enumerate} From owner-pbwg-comm@CS.UTK.EDU Wed Jun 2 18:26:16 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA13296; Wed, 2 Jun 93 18:26:16 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02392; Wed, 2 Jun 93 18:26:01 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 2 Jun 1993 18:26:00 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sp2.csrd.uiuc.edu by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA02372; Wed, 2 Jun 93 18:25:53 -0400 Received: from sp94.csrd.uiuc.edu.csrd.uiuc.edu (sp94.csrd.uiuc.edu) by sp2.csrd.uiuc.edu with SMTP id AA25307 (5.67a/IDA-1.5); Wed, 2 Jun 1993 17:25:45 -0500 Received: by sp94.csrd.uiuc.edu.csrd.uiuc.edu (4.1/SMI-4.1) id AA03416; Wed, 2 Jun 93 17:25:43 CDT Date: Wed, 2 Jun 93 17:25:43 CDT From: schneid@csrd.uiuc.edu (David John Schneider) Message-Id: <9306022225.AA03416@sp94.csrd.uiuc.edu.csrd.uiuc.edu> To: R.Hockney@pac.soton.ac.uk Cc: pbwg-comm@cs.utk.edu, perfect.steering@csrd.uiuc.edu In-Reply-To: <18087.9306021659@calvados.pac.soton.ac.uk> (R.Hockney@pac.soton.ac.uk) Subject: Re: Revised SPEEDUP section Several weeks ago at the first meeting of the newly reconstituted SPEC-Perfect steering committee, we also got tangled up in policy questions related to speedup and other derived quantities. After the usual heated discussions, we decided to simply report only elapsed times (i.e. time-to-solution or it's inverse). As usual, this conclusion was a result of a lot of discussion, much of which I expect was the same as occurred at the PBWG meeting. Since the SPEC-Perfect meeting, I have spent some time trying to understand why speedup and other issues always provoke such contentious discussions. The current debate in the PBWG has prompted me to try to write down some of these ideas to clarify my own thinking. The basic conclusion that I have reached is that there are two important classes of problems in basic performance evaluation methodology today, both of which can be readily addressed. Since I have been unable to attend PBWG meetings and directly participate in methodology subcommittee meetings, I'll attempt to summarize my thoughts here. The first class of problems is very basic -- we are searching for something which does not exist. For example, there is tendency to expend considerable effort to define a single, universal figure of merit for comparing machines when, in fact, there are good reason to think that this cannot ever be done in a unique and unbiased manner. It would be nice to be able to say that machine X is better (eg. faster) than machine Y, with no qualifications. However, this simply isn't possible (see below). Second, there are other performance measures such as speedup which are, at best, loosely correlated with an end user's perception of delivered performance and therefore are of little real utility. I agree with all of Roger's comments in this regard. Nevertheless, these loosely defined performance measures tend to get a lot of press in both academic and popular publications because they are "easy to understand". In fact, the contentious debates on these issues indicate that this "understanding" is largely illusory and unsatisfactory when taken out of the context or when this context is not provided at all. I feel that both of these classes of problems can be overcome by adopting and adhering to an axiomatic approach. The utility an axiomatic approach in defining a set of logically consistent performance measures has been recently advocated by directly by David Snelling. Others such as Roger Hockney, John Larson, David Bailey have argued forcefully for the need of a well defined mathematical framework, and the axiomatic approach provides one alternative for constructing such a framework. One of the the most important aspects of axiomatic approach is that it forces one to precisely and explicitly state assumptions regarding the measurement process that have previously remained implicit or imprecise. In the axiomatic approach, computer performance evaluation becomes the study of the complicated mapping from the set of computer codes into the multidimensional space of measureable quantities which obey the prescribed axioms (additivity, positivity, etc.). A second important aspect of the axiomatic approach is that makes it clear that one cannot construct a unique, universal, unbiased ranking scheme for computer performance simply because there is no natural ordering to the underlying space of measurements. Therefore, the search for a universal figure of merit is certain to fail. Would a well informed computer scientist attempt to develop an algorithm for determining whether or not a Turing machine will halt when presented with an arbitrary input? Would a practically minded physicist or engineer write a grant to develop a perpetual motion machine or a faster-than-light communication system? In all cases, if one accepts the usual mathematical description of the underlying problem, then all of these endeavors can be proven futile. Why then in benchmarking do we continue to pursue the goal of a single figure of merit? As emphasized by David Snelling, the axiomatic approach has been extrordinarily successful in the physical sciences. For example, one can meaningfully ascribe quantities such as energy and momentum to particles, waves fields, and combinations thereof. More recently, in the hands of Shannon and others the field of communication theory underwent a spectacular revolution from a raw empiricism to a quantitative science in a space of only several years by employing a very simple system of axioms and a clear set of definitions. In this case the basic axioms were largely borrowed from the foundations of statistical physics (information content or entropy should be an extensive, non-negative function, etc.). Shannon's major contibution was to recognize that it is possible distinguish between the precisely definable notion of "information content", and the loosely defined "meaning" of this information to the recipient of the message. I believe that the same type of precise thinking which was so successfully applied by Shannon to transform communication theory into a quantitative science also needs to be applied to performance evaluation. The list of existing performance measures which can be incorporated into an axiomatic framework is very remarkably short, and Roger Hockney has enumerated essentially of the time-based measures in hsi previous versions of PBWG documents. I personally think it is a large step backwards to include speedup in light of the sound arguments against it. The SPEC-Perfect group steering committee concluded that the major argument in favor of speedup and single figures of merit was their current popularity, and this reason was insufficient for us to adopt them as part of our basic measurement and reporting methodology. -- Dave Schneider University of Illinois at Urbana-Champaign Center for Supercomputing Research and Development 367 Computer and Systems Research Laboratory 1308 W. Main Street Urbana, IL 61801-2307 MC-264 phone : (217) 244-0055 fax : (217) 244-1351 E-mail: schneid@csrd.uiuc.edu ======================================================================== Return-Path: Errors-To: owner-pbwg-comm@CS.UTK.EDU X-Resent-To: pbwg-comm@CS.UTK.EDU ; Wed, 2 Jun 1993 14:21:19 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU From: R.Hockney@pac.soton.ac.uk Date: Wed, 2 Jun 93 16:59:55 GMT To: pbwg-comm@cs.utk.edu Subject: Revised SPEEDUP section Because the definition of Speedup is of such general concern to all members, not just those in the Methodology subcommittee, I am sending this proposed ammendment to all committee members: *********************** \subsection{Speedup and Efficiency} Speedup is a popular metric that has been used for many years to compare the performance of parallel computers. However its definition is open to ambiguity and misuse because it always begs the question "Speedup over what?"; a question that is often not clearly answered in publications using this metric. Whilst preferring the use of absolute measures of performance, such as Benchmark Performance defined earlier, the PARKBENCH committee accepted that speedup would probably continue to be used, and that the best policy was to sharpen-up its definition. Speedup is universally defined as \begin{equation} frac{T_1}{T_p} \end{equation} where $T_p$ is the p-processor time to perform some benchmark, and $T_1$ is called the one-processor time. There is no doubt about the meaning of $T_p$, because this is measured time $T(N;p)$ to perform the benchmark. There is, however, usually considerable discussion over the meaning of $T_1$, [stuff deleted] From owner-pbwg-comm@CS.UTK.EDU Thu Jun 3 11:43:03 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA17970; Thu, 3 Jun 93 11:43:03 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10097; Thu, 3 Jun 93 11:42:35 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Thu, 3 Jun 1993 11:42:34 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sp2.csrd.uiuc.edu by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10084; Thu, 3 Jun 93 11:42:31 -0400 Received: by sp2.csrd.uiuc.edu id AA04316 (5.67a/IDA-1.5); Thu, 3 Jun 1993 10:42:03 -0500 Date: Thu, 3 Jun 1993 10:42:03 -0500 From: "John L. Larson" Message-Id: <199306031542.AA04316@sp2.csrd.uiuc.edu> To: pbwg-comm@cs.utk.edu, perfect.steering@csrd.uiuc.edu Subject: recent papers I am sending you postscript versions of two recent papers for your information. The first one which includes performance metric definitions will appear in the August issue of the Proceedings of the IEEE. The second one describes a workload characterization study and has been submitted to Supercomputing '93. Any comments are appreciated. thanks john From owner-pbwg-comm@CS.UTK.EDU Thu Jun 3 11:44:03 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA17985; Thu, 3 Jun 93 11:44:03 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10266; Thu, 3 Jun 93 11:44:22 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Thu, 3 Jun 1993 11:44:18 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sp2.csrd.uiuc.edu by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10228; Thu, 3 Jun 93 11:44:07 -0400 Received: by sp2.csrd.uiuc.edu id AA04336 (5.67a/IDA-1.5); Thu, 3 Jun 1993 10:43:34 -0500 Date: Thu, 3 Jun 1993 10:43:34 -0500 From: "John L. Larson" Message-Id: <199306031543.AA04336@sp2.csrd.uiuc.edu> To: pbwg-comm@cs.utk.edu, perfect.steering@csrd.uiuc.edu Subject: IEEE paper %!PS-Adobe-2.0 %%Creator: dvips 5.47 Copyright 1986-91 Radical Eye Software %%Title: /homes/bradley/ieee.dvi %%Pages: 27 1 %%BoundingBox: 0 0 612 792 %%EndComments %%BeginProcSet: texc.pro /TeXDict 200 dict def TeXDict begin /N /def load def /B{bind def}N /S /exch load def /X{S N}B /TR /translate load N /isls false N /vsize 10 N /@rigin{ isls{[0 1 -1 0 0 0]concat}if 72 Resolution div 72 VResolution div neg scale Resolution VResolution vsize neg mul TR matrix currentmatrix dup dup 4 get round 4 exch put dup dup 5 get round 5 exch put setmatrix}N /@letter{/vsize 10 N}B /@landscape{/isls true N /vsize -1 N}B /@a4{/vsize 10.6929133858 N}B /@a3{ /vsize 15.5531 N}B /@ledger{/vsize 16 N}B /@legal{/vsize 13 N}B /@manualfeed{ statusdict /manualfeed true put}B /@copies{/#copies X}B /FMat[1 0 0 -1 0 0]N /FBB[0 0 0 0]N /nn 0 N /IE 0 N /ctr 0 N /df-tail{/nn 8 dict N nn begin /FontType 3 N /FontMatrix fntrx N /FontBBox FBB N string /base X array /BitMaps X /BuildChar{CharBuilder}N /Encoding IE N end dup{/foo setfont}2 array copy cvx N load 0 nn put /ctr 0 N[}B /df{/sf 1 N /fntrx FMat N df-tail} B /dfs{div /sf X /fntrx[sf 0 0 sf neg 0 0]N df-tail}B /E{pop nn dup definefont setfont}B /ch-width{ch-data dup length 5 sub get}B /ch-height{ch-data dup length 4 sub get}B /ch-xoff{128 ch-data dup length 3 sub get sub}B /ch-yoff{ ch-data dup length 2 sub get 127 sub}B /ch-dx{ch-data dup length 1 sub get}B /ch-image{ch-data dup type /stringtype ne{ctr get /ctr ctr 1 add N}if}B /id 0 N /rw 0 N /rc 0 N /gp 0 N /cp 0 N /G 0 N /sf 0 N /CharBuilder{save 3 1 roll S dup /base get 2 index get S /BitMaps get S get /ch-data X pop /ctr 0 N ch-dx 0 ch-xoff ch-yoff ch-height sub ch-xoff ch-width add ch-yoff setcachedevice ch-width ch-height true[1 0 0 -1 -.1 ch-xoff sub ch-yoff .1 add]/id ch-image N /rw ch-width 7 add 8 idiv string N /rc 0 N /gp 0 N /cp 0 N{rc 0 ne{rc 1 sub /rc X rw}{G}ifelse}imagemask restore}B /G{{id gp get /gp gp 1 add N dup 18 mod S 18 idiv pl S get exec}loop}B /adv{cp add /cp X}B /chg{rw cp id gp 4 index getinterval putinterval dup gp add /gp X adv}B /nd{/cp 0 N rw exit}B /lsh{rw cp 2 copy get dup 0 eq{pop 1}{dup 255 eq{pop 254}{dup dup add 255 and S 1 and or}ifelse}ifelse put 1 adv}B /rsh{rw cp 2 copy get dup 0 eq{pop 128}{dup 255 eq{pop 127}{dup 2 idiv S 128 and or}ifelse}ifelse put 1 adv}B /clr{rw cp 2 index string putinterval adv}B /set{rw cp fillstr 0 4 index getinterval putinterval adv}B /fillstr 18 string 0 1 17{2 copy 255 put pop}for N /pl[{adv 1 chg}bind{adv 1 chg nd}bind{1 add chg}bind{1 add chg nd}bind{adv lsh}bind{ adv lsh nd}bind{adv rsh}bind{adv rsh nd}bind{1 add adv}bind{/rc X nd}bind{1 add set}bind{1 add clr}bind{adv 2 chg}bind{adv 2 chg nd}bind{pop nd}bind]N /D{ /cc X dup type /stringtype ne{]}if nn /base get cc ctr put nn /BitMaps get S ctr S sf 1 ne{dup dup length 1 sub dup 2 index S get sf div put}if put /ctr ctr 1 add N}B /I{cc 1 add D}B /bop{userdict /bop-hook known{bop-hook}if /SI save N @rigin 0 0 moveto}N /eop{clear SI restore showpage userdict /eop-hook known{eop-hook}if}N /@start{userdict /start-hook known{start-hook}if /VResolution X /Resolution X 1000 div /DVImag X /IE 256 array N 0 1 255{IE S 1 string dup 0 3 index put cvn put}for}N /p /show load N /RMat[1 0 0 -1 0 0]N /BDot 260 string N /rulex 0 N /ruley 0 N /v{/ruley X /rulex X V}B /V statusdict begin /product where{pop product dup length 7 ge{0 7 getinterval (Display)eq}{pop false}ifelse}{false}ifelse end{{gsave TR -.1 -.1 TR 1 1 scale rulex ruley false RMat{BDot}imagemask grestore}}{{gsave TR -.1 -.1 TR rulex ruley scale 1 1 false RMat{BDot}imagemask grestore}}ifelse B /a{moveto}B /delta 0 N /tail{dup /delta X 0 rmoveto}B /M{S p delta add tail}B /b{S p tail} B /c{-4 M}B /d{-3 M}B /e{-2 M}B /f{-1 M}B /g{0 M}B /h{1 M}B /i{2 M}B /j{3 M}B /k{4 M}B /w{0 rmoveto}B /l{p -4 w}B /m{p -3 w}B /n{p -2 w}B /o{p -1 w}B /q{p 1 w}B /r{p 2 w}B /s{p 3 w}B /t{p 4 w}B /x{0 S rmoveto}B /y{3 2 roll p a}B /bos{ /SS save N}B /eos{clear SS restore}B end %%EndProcSet TeXDict begin 1000 300 300 @start /Fa 36 122 df<127012F8A3127005057C840D>46 D48 DIII<1306A2 130EA2131E132EA2134E138EA2EA010E1202A212041208A212101220A2124012C0B512F038000E 00A7EBFFE0141E7F9D17>II<137CEA0182EA0701380E0380EA0C07121C3838030090C7FC12781270A2EAF1F0EAF21CEA F406EAF807EB0380A200F013C0A51270A214801238EB07001218EA0C0E6C5AEA01F0121F7E9D17 >I<1240387FFFC01480A238400100EA8002A25B485AA25B5BA25BA213C0A212015B1203A41207 A76CC7FC121F7D9D17>III<5B497EA3497EA3EB09E0A3EB10F0A3EB2078A3497EA3497E A2EBFFFE3801000FA30002EB0780A348EB03C0120E001FEB07E039FFC03FFE1F207F9F22>65 DI<90380FC04090387030C03801C00938038005 38070003000E1301001E1300121C123C007C1440A2127800F81400A91278007C1440A2123C121C 001E1480120E6CEB0100380380026C6C5A38007038EB0FC01A217D9F21>IIII<90 380FE02090387818609038E004E03803800238070001481300001E1460A25A1520127C127800F8 1400A7EC7FFCEC03E000781301127C123CA27EA27E7E380380023900E00460903878182090380F E0001E217D9F24>I77 D<39FF803FF83907C007C0EC03803905E00100A2EA04F01378A2133CA213 1E130FA2EB0781A2EB03C1EB01E1A2EB00F1A21479143DA2141FA28080A2000E7F121F38FFE001 1D1F7E9E22>I82 D<3807E080EA0C19EA1007EA3003EA6001A212E01300A36C1300A21278127FEA3FF0EA1FFC6C7E EA03FF38001F801307EB03C0A213011280A400C01380130300E01300EAF006EACE0CEA81F81221 7D9F19>I<39FFF003FF391F8000F8000F1460000714407F6C6C138012019038F0010000005BEB F802EB7C04133CEB3E08131EEB1F10EB0FB0EB07A014C01303AB1307EB7FFE201F7F9E22>89 D97 D<120E12FE120EAA133EEBC380380F01C0EB00E0120E1470 A21478A61470A214E0120F380D01C0380CC300EA083E15207F9F19>IIII<137C13C6EA018F1203EA07061300A7EAFFF0EA0700B2 EA7FF01020809F0E>I<14E03803E330EA0E3CEA1C1C38380E00EA780FA5EA380E6C5AEA1E38EA 33E00020C7FCA21230A2EA3FFE381FFF806C13C0383001E038600070481330A4006013606C13C0 381C03803803FC00141F7F9417>I<390E1F01F039FE618618390E81C81C390F00F00EA2000E13 E0AE3AFFE7FE7FE023147F9326>109 DI114 DI<38FF83F8381E01E0381C00C06C1380A338070100A2EA 0382A3EA01C4A213ECEA00E8A21370A3132015147F9318>118 D<38FF83F8381E01E0381C00C0 6C1380A338070100A2EA0382A3EA01C4A213ECEA00E8A21370A31320A25BA3EAF080A200F1C7FC 1262123C151D7F9318>121 D E /Fb 15 119 df<1430146014C0EB0180EB03005B130E130C5B 1338133013705B5B12015B1203A290C7FC5A1206120EA2120C121CA312181238A45AA75AB3A312 70A77EA41218121CA3120C120EA2120612077E7FA212017F12007F13701330133813187F130E7F 7FEB0180EB00C014601430146377811F>18 D<12C012607E7E7E120E7E7E6C7E7F12007F137013 3013381318131CA2130C130E13061307A27F1480A3130114C0A4EB00E0A71470B3A314E0A7EB01 C0A414801303A314005BA21306130E130C131CA213181338133013705B5B12015B48C7FC5A120E 120C5A5A5A5A14637F811F>I<14181430146014E014C0EB01801303EB07001306130E130C131C 5BA25BA25BA212015BA2485AA3120790C7FCA25A120EA2121EA3121CA2123CA412381278A81270 12F0B3A812701278A81238123CA4121CA2121EA3120EA2120F7EA27F1203A36C7EA27F1200A213 70A27FA27F130C130E13061307EB03801301EB00C014E0146014301418157C768121>32 D<12C012607E123812187E120E7E7E7F12017F6C7EA21370A27FA2133C131CA27FA3130F7FA214 801303A214C0A31301A214E0A4130014F0A814701478B3A8147014F0A814E01301A414C0A21303 A31480A213071400A25B130EA35BA2133C1338A25BA25BA2485A5B120390C7FC5A120E120C5A12 3812305A5A157C7F8121>I<1318137813F0EA01E0EA03C0EA0780EA0F005A121E123E123C127C A2127812F8B3A50D25707E25>56 D<12C012F0127C121E7EEA078013C01203EA01E013F0120013 F8A3137CB3A50E25797E25>I<12F8B3A51278127CA2123C123E121E121F7EEA0780EA03C0EA01 E0EA00F0137813180D25708025>I<137CB3A513F8A313F0120113E0EA03C012071380EA0F0012 1E127C12F012C00E25798025>I<137CB3A613F8A313F0120113E0120313C0EA07801300120E5A 5A12F012C012F012387E7E7E1380EA03C013E0120113F0120013F8A3137CB3A60E4D798025>I< 12F8B3A61278127CA2123CA2123E121E7EA2EA0780EA03C01201EA00E013781318137813E0EA01 C01203EA0780EA0F00A2121E123E123CA2127CA2127812F8B3A60D4D708025>I<12F8AE050E70 8025>I88 D116 D<12C0B3A9021B64802C>I<387FFF80B5FC00C0C7FCB3A611 1A64812C>I E /Fc 9 116 df<0007B512803800E003EC0100A3EA01C0A21440A248485A138113 FF1381D80701C7FCA390C8FC120EA45AEAFFC019177F9616>70 D99 D101 D<130E1313133713361360A5EA07 FCEA00C0A5EA0180A6EA0300A4126612E65A1278101D7E9611>I<120313801300C7FCA6121C12 241246A25A120C5AA31231A21232A2121C09177F960C>105 D<1318133813101300A6EA01C0EA 0220EA0430A2EA08601200A313C0A4EA0180A4EA630012E312C612780D1D80960E>I<38383C1E 3844C6633847028138460301388E0703EA0C06A21406EA180C1520140C154038301804EC07801B 0E7F8D1F>109 DI115 D E /Fd 1 121 df<121FEA3080EA7040EA6060EAE0E0A21340 13001260127012307E121C1233EA7180EA61C012E013E0A31260EA70C01231EA1980EA07007EEA 018013C0120013E0124012E0A2EAC0C01241EA2180EA1F000B257D9C12>120 D E /Fe 52 123 df<903901FF81FE011F9038EFFF80903A7F80FF87C0903AFC00FE0FE03801F8 01000314FCEA07F0EE07C093C7FCA7B712F8A32707F001FCC7FCB3A33A7FFF1FFFE0A32B2A7FA9 28>11 D<903A01FF803FE0011F9038E3FFF8903A7F80FFF01E903AFE007F801FD801F849485A00 0349EC7F80D807F05BA30200EC3F00171E94C7FCA4B91280A33B07F000FE003FB3A33C7FFF0FFF E3FFF8A3352A7FA939>14 D<121C127FA2EAFF8013C0A2127FA2121C1200A2EA0180A3EA0300A2 12065A5A5A12200A157BA913>39 D45 D<121C123E127FEAFF80A3EA7F 00123E121C09097B8813>I<130E131E137EEA07FE12FFA212F81200B3AB387FFFFEA317277BA6 22>49 DII<140E141E143E147E14FEA213011303EB077E130EA2131C13381370 13E0A2EA01C0EA0380EA0700120EA25A5A5A5AB612F8A3C7EAFE00A890387FFFF8A31D277EA622 >I<000C1303380F803FEBFFFE5C5C5C5C5C49C7FC000EC8FCA6EB7FC0380FFFF8EB80FC380E00 3E000C133FC7EA1F8015C0A215E0A21218127C12FEA315C05A0078EB3F80A26CEB7F00381F01FE 380FFFF800035BC613801B277DA622>II<1238123E003FB512F0A34814E015C0158015003870000EA25C485B5C5CC7FC495A495A 1307A249C7FCA25BA25B133EA2137EA413FEA8137C13381C297CA822>I<121C123E127FEAFF80 A3EA7F00123E121CC7FCA9121C123E127FEAFF80A3EA7F00123E121C091B7B9A13>58 D65 DI<91393FF0018090 3903FFFE03010FEBFF8790393FF007DF9039FF8001FF4848C7127F4848143FD807F0141F000F15 0F48481407A2485A1603127F5B93C7FC12FFA9127FA26DEC0380123FA26C7EEE07006C7E000715 0ED803FC141E6C6C5C6C6C6C13F890393FF007E0010FB55A010391C7FC9038003FF829297CA832 >III<91387FE003903903FFFC07011FEBFF0F90393FF00FFF9038FF80014848C7FCD8 03F8143F485A000F81484880A2485A82127F5B93C7FC12FFA84AB512F8127FA26DC71300123FA2 6C7EA26C7E12076C7EEA01FE6C6C6C5A90393FF007BF6DB5121F0103497E9039007FF0032D297C A836>71 DI76 DI80 DII<90387F80603903FFF0E04813F9380F807F381F 001F003E1307481303140112FCA214007EA26C140013C0EA7FFEEBFFE06C13FC6C7F6CEBFF806C 14C06C14E0C6FC010713F0EB007FEC0FF8140714030060130112E0A36C14F0A26C13036C14E0B4 EB07C09038E01F8000F3B5120000E05B38C01FF01D297CA826>I<007FB712C0A39039803FC03F D87E00140700781503A20070150100F016E0A2481500A5C71500B3A4017FB512E0A32B287EA730 >II 89 D<48B47E000713F0380F81F8381FC07EA280D80F801380EA0700C7FCA3EB0FFF90B5FC3807 FC3FEA0FE0EA3F8013005A12FEA4007E137F007F13DF393F839FFC380FFF0F3801FC031E1B7E9A 21>97 DIIIII<90 38FF81F00003EBE7F8390FC1FE7C381F80FC9038007C3848EB7E1048EB7F00A66C137E6C137CEB 80FC380FC1F8381FFFE0001813800038C8FCA2123C123E383FFFF86C13FF15806C14C06C14E000 1F14F0383E000748EB01F8481300A4007CEB01F0003C14E0001FEB07C0390FC01F803903FFFE00 38007FF01E287E9A22>II<1207EA0F80EA1FC0EA3FE0A3EA1FC0EA0F80EA07 00C7FCA7EAFFE0A3120FB3A3EAFFFEA30F2B7DAA14>I107 DI<3BFFC0 7F800FF0903AC1FFE03FFC903AC383F0707E3B0FC603F8C07F903ACC01F9803F01D8D9FF001380 01F05BA201E05BB03CFFFE1FFFC3FFF8A3351B7D9A3A>I<38FFC07F9038C1FFC09038C787E039 0FCE03F013D88113F0A213E0B03AFFFE3FFF80A3211B7D9A26>II<38FFE1FE9038E7FF809038FE07E039 0FF803F0496C7E496C7E818181A21680A716005DA25D4A5A01F05B6D485A9038FE0FE09038E7FF 80D9E1FCC7FC01E0C8FCA9EAFFFEA321277E9A26>I<38FFC1F0EBC7FCEBCE3E380FD87FA213F0 143E141CEBE000B0B5FCA3181B7E9A1C>114 D<3803FE30380FFFF0EA1E03EA380048137012F0 A27E6C1300EAFFE0EA7FFEEBFF806C13E06C13F0000713F8C6FCEB03FC13000060137C00E0133C 7E14387E6C137038FF01E038F7FFC000C11300161B7E9A1B>I<1370A413F0A312011203A21207 381FFFF0B5FCA23807F000AD1438A61203EBF870000113603800FFC0EB1F8015267FA51B>I<39 FFE03FF8A3000F1303B214071207140F3A03F03BFF803801FFF338003FC3211B7D9A26>I<3BFF FE7FFC0FFEA33B0FE007E000E03B07F003F001C0A29039F807F80300031680A23B01FC0EFC0700 A2D9FE1E5B000090381C7E0EA29039FF383F1E017F141C0278133C90393FF01FB8A216F86D486C 5AA26D486C5AA36D486C5AA22F1B7F9A32>119 D<39FFFC0FFFA33907F003C06C6C485AEA01FC 6C6C48C7FCEBFF1E6D5AEB3FF86D5A130FA2130780497E497E131EEB3C7F496C7E496C7ED801E0 7FEBC00F00036D7E3AFFF01FFF80A3211B7F9A24>I<3AFFFE03FF80A33A07F0007000A26D13F0 00035CEBFC0100015CA26C6C485AA2D97F07C7FCA2148FEB3F8E14DEEB1FDCA2EB0FF8A36D5AA2 6D5AA26D5AA2495AA2EA3807007C90C8FCEAFE0F130E131E5BEA7C78EA3FE0EA0FC021277F9A24 >I<003FB51280A29038007F00003C13FEEA3801387803FC5CEA7007495A5CC6485A133F495A91 C7FC5B3901FE038013FCEA03F81207380FF00713E0001F140048485A495A387F007FB6FCA2191B 7E9A1F>I E /Ff 15 121 df0 D<6C13026C13060060130C6C13186C13 306C13606C13C03803018038018300EA00C6136C1338A2136C13C6EA018338030180380600C048 136048133048131848130C4813064813021718789727>2 D15 D<150C153C15F0EC03C0EC0F0014 3C14F0EB07C0011FC7FC1378EA01E0EA0780001EC8FC127812E01278121EEA0780EA01E0EA0078 131FEB07C0EB00F0143C140FEC03C0EC00F0153C150C1500A8007FB512F8B612FC1E277C9F27> 20 D<487E48CAFCA21206A25A5A1270B812C0A20070CAFC12187E7EA27EA26C7E2A127C9432> 32 D<166082A282A28282EE0380B812E017C0C9EA0380EE06005E5EA25EA25E2B127D9432>I<13 C0B3B100C013C0EAF0C33838C700EA1CCEEA06D8EA03F06C5AA26C5A1340122D7DA219>35 D<12C07E12707E7E7E7E6C7E6C7E6C7E13707F7F7F7F6D7E6D7E6D7E1470808080806E7E6E7E6E 7E15708181816F13209238038060ED01C0ED00E0EE70C01638161C160E16071603163F923801FF E0923803C0602B2B7DA032>38 D<176017E0EE01C0EE0380EE0700160E5E5E5E5E4B5A4B5A4BC7 FC150E5D5D5D5D4A5A4A5A4AC8FC140E5C5C5C5C495A495A49C9FC130E6C5A6C5A5B5BEA61C0EA 63800067CAFC126E127C1278EA7F80EAFFF0EAC0782B2B7DA032>46 D<134013C0A2EA0180A3EA 0300A31206A25AA35AA35AA35AA35AA41260A37EA37EA37EA37EA27EA3EA0180A3EA00C0A21340 0A327BA413>104 D<12C0A31260A37EA37EA27EA37EA37EA3EA0180A3EA00C0A4EA0180A3EA03 00A31206A35AA35AA25AA35AA35AA30A327DA413>I<12C0B3B3AD02317AA40E>II<160116031606A2160CA21618A21630A21660A216C0A2ED0180A2 ED0300A21506A25DA25DA25D1206001E5C122F004F5CEA87800007495AEA03C04AC7FCA23801E0 06A26C6C5AA2EB7818A26D5AA26D5AA26D5AA26D5AA26DC8FCA228327D812A>112 D120 D E /Fg 56 123 df11 D<137E3801C180EA0301380703C0120EEB018090C7FC A5B512C0EA0E01B0387F87F8151D809C17>II34 D<13401380EA0100120212065AA25AA25AA212701260A312E0AC1260A312 701230A27EA27EA27E12027EEA008013400A2A7D9E10>40 D<7E12407E7E12187EA27EA27EA213 801201A313C0AC1380A312031300A21206A25AA25A12105A5A5A0A2A7E9E10>I<126012F0A212 701210A41220A212401280040C7C830C>44 DI<126012F0A212600404 7C830C>I48 D<12035A123F12C71207B3A4EA0F80EAFFF80D1C7C9B 15>III<130CA2131C133CA2135C13DC13 9CEA011C120312021204120C1208121012301220124012C0B512C038001C00A73801FFC0121C7F 9B15>II<13F0EA030CEA0604EA0C0E EA181E1230130CEA7000A21260EAE3E0EAE430EAE818EAF00C130EEAE0061307A51260A2EA7006 EA300E130CEA1818EA0C30EA03E0101D7E9B15>I<1240387FFF801400A2EA4002485AA25B485A A25B1360134013C0A212015BA21203A41207A66CC7FC111D7E9B15>I<1306A3130FA3EB1780A3 EB23C0A3EB41E0A3EB80F0A200017FEB0078EBFFF83803007C0002133CA20006133E0004131EA2 000C131F121E39FF80FFF01C1D7F9C1F>65 DI<9038 1F8080EBE0613801801938070007000E13035A14015A00781300A2127000F01400A80070148012 78A212386CEB0100A26C13026C5B380180083800E030EB1FC0191E7E9C1E>IIII73 D77 D79 DI<3807E080EA1C19EA3005EA7003EA600112E01300A36C13007E127CEA7FC0 EA3FF8EA1FFEEA07FFC61380130FEB07C0130313011280A300C01380A238E00300EAD002EACC0C EA83F8121E7E9C17>83 D<007FB512C038700F010060130000401440A200C014201280A3000014 00B1497E3803FFFC1B1C7F9B1E>I<397FF0FFC0390FC03E0038078018EA03C0EBE01000015BEB F06000001340EB7880137D013DC7FC7F131F7F80A2EB13C0EB23E01321EB41F0EBC0F8EB807838 01007C48133C00027F0006131F001FEB3F8039FFC0FFF01C1C7F9B1F>88 D<39FFF007FC390F8001E00007EB0080EBC00100031400EBE002EA01F000005B13F8EB7808EB7C 18EB3C106D5A131F6D5A14C06D5AABEB7FF81E1C809B1F>I92 D97 D<12FC121CAA137CEA1D86EA1E03381C018014C0130014E0A614C013011480381E0300EA1906EA 10F8131D7F9C17>II<133F1307AAEA03E7EA0C17EA180F487E1270126012E0A6126012 7012306C5AEA0C373807C7E0131D7E9C17>II<13F8EA018CEA071E12 06EA0E0C1300A6EAFFE0EA0E00B0EA7FE00F1D809C0D>II<12FC121CAA137C1387EA1D0300 1E1380121CAD38FF9FF0141D7F9C17>I<1218123CA21218C7FCA712FC121CB0EAFF80091D7F9C 0C>I<12FC121CAAEB3FC0EB0F00130C13085B5B5B13E0121DEA1E70EA1C781338133C131C7F13 0F148038FF9FE0131D7F9C16>107 D<12FC121CB3A9EAFF80091D7F9C0C>I<39FC7E07E0391C83 8838391D019018001EEBE01C001C13C0AD3AFF8FF8FF8021127F9124>IIII< EAFCE0EA1D30EA1E78A2EA1C301300ACEAFFC00D127F9110>114 DI<1204A4120CA2121C123CEAFFE0EA1C00A91310A5120CEA0E20EA03C00C1A7F9910>I<38 FC1F80EA1C03AD1307120CEA0E1B3803E3F014127F9117>I<38FF07E0383C0380381C0100A2EA 0E02A26C5AA3EA0388A213D8EA01D0A2EA00E0A3134013127F9116>I<39FF3FCFE0393C0F0380 381C07011500130B000E1382A21311000713C4A213203803A0E8A2EBC06800011370A2EB803000 0013201B127F911E>I<387F8FF0380F03801400EA0702EA0384EA01C813D8EA00F01370137813 F8139CEA010E1202EA060738040380381E07C038FF0FF81512809116>I<38FF07E0383C038038 1C0100A2EA0E02A26C5AA3EA0388A213D8EA01D0A2EA00E0A31340A25BA212F000F1C7FC12F312 66123C131A7F9116>II E /Fh 5 54 df<120C121C12EC120CAF EAFFC00A137D9211>49 D<121FEA60C01360EAF07013301260EA0070A2136013C012011380EA02 005AEA08101210EA2020EA7FE012FF0C137E9211>II<136013E0A2EA 016012021206120C120812101220126012C0EAFFFCEA0060A5EA03FC0E137F9211>II E /Fi 60 123 df12 D<91390FF01FE091393838601891396079C01C9139C07B803C0101EB330003071338D903801400 A2150EA2EB0700A35D90B712E090390E001C00A2EE01C01538A25BEE0380A21570A2EE07005BA2 03E01308EE0E10A25B1720913801C006EE03C0016091C7FC01E05B140313C092C8FCEA71C738F1 8F06EB0F0C38620618383C03E02E2D82A22B>14 D<1480EB010013025B5B5B13305B5BA2485A48 C7FCA21206A2120E120C121C1218A212381230A21270A21260A212E0A35AAD12401260A2122012 3012107E113278A414>40 D<13087F130613021303A27F1480AD1303A31400A25BA21306A2130E 130CA2131C131813381330A25BA25B485AA248C7FC120612045A5A5A5A5A113280A414>I45 D<127012F8A212F012E005057A840F>I<150815181530A2156015C0 A2EC0180A2EC03001406A25CA25C5CA25CA25C495AA249C7FCA21306A25B5BA25BA25B5BA2485A A248C8FC1206A25AA25A5AA25AA25AA25A1D317FA419>II<13011303A21306131E132EEA03CEEA001CA41338A4 1370A413E0A4EA01C0A4EA0380A41207EAFFFC10217AA019>II<14181438A21470A314E0A314C01301148013031400A21306A25BA25B1310EB 3180EB61C0EB438013831201EA03033802070012041208EA3FC7EA403E38800FF038000E00A25B A45BA31330152B7EA019>52 D54 D<380278023804FC041207000F1308EB0E18381E 0670381803A03830006000201340006013C000401380EA8001EA000314005B1306130EA25BA213 3C13381378A25BA3485AA312035BA26C5A172279A019>I<1207EA0F80A21300120EC7FCAB1270 12F8A25A5A09157A940F>58 D<001FB6FCA2C9FCA8B612F8A2200C7B9125>61 D<1403A25CA25CA25C142FA2144F15801487A2EB01071302A21304A21308A2131013301320EB7F FF90384007C013801403EA01005A12025AA2120C003C1307B4EB3FFC1E237DA224>65 D<027F138090390380810090380E00630138132749131F49130E485A485A48C7FC481404120E12 1E5A5D4891C7FCA35AA55A1520A25DA26C5C12704AC7FC6C130200185B001C5B00061330380381 C0D800FEC8FC212479A223>67 D<90B512F090380F003C150E81011EEB0380A2ED01C0A25B16E0 A35BA449EB03C0A44848EB0780A216005D4848130E5D153C153848485B5D4A5A0207C7FC000F13 1CB512F023227DA125>I<90B6128090380F00071501A2131EA21600A25BA2140192C7FCEB7802 A21406140EEBFFFCEBF00CA33801E008A21504EC0008485AA25DA248485B15605D1401000F1307 B65A21227DA121>I<90B6FC90380F000F1503A2131EA21502A25BA214011500EB7802A2140614 0EEBFFFCEBF00CA33801E008A391C7FC485AA4485AA4120FEAFFFC20227DA120>I<027F138090 390380810090380E00630138132749131F49130E485A485A48C7FC481404120E121E5A5D4891C7 FCA35AA4EC3FFC48EB01E0A34A5AA27E12704A5A7E0018130F001C131300060123C7FC380381C1 D800FEC8FC212479A226>I<9039FFF87FFC90390F000780A3011EEB0F00A449131EA4495BA490 B512F89038F00078A348485BA44848485AA44848485AA4000F130739FFF87FFC26227DA124>I< EBFFF8EB0F00A3131EA45BA45BA45BA4485AA4485AA4485AA4120FEAFFF815227DA113>I<9038 07FFC09038003C00A35CA45CA4495AA4495AA4495AA449C7FCA212381278EAF81EA2485AEA4038 5BEA21E0EA1F801A237CA11A>I76 DI<01FFEB0FFC90390F8001E016 80ECC0000113EB0100A2EB11E0A201211302EB20F0A39038407804A3143C01805B143E141EA239 01001F10140FA2EC0790000214A0A2EC03E0A2485C1401A2120C001E6D5AEAFFC026227DA124> I<14FE903807838090380C00E0013813704913385B4848131C485A48C7FC48141E121E121C123C A25AA348143CA31578A25A15F0A2EC01E015C06C1303EC0780007014000078130E00385B6C5B6C 13E038070380D801FCC7FC1F2479A225>I<90B512E090380F0038151E150E011E1307A449130F A3151E5B153C157815E09038F003C09038FFFE0001F0C7FCA2485AA4485AA4485AA4120FEAFFF8 20227DA121>I<14FE903807838090380C00E0013813704913385B4848133C4848131C48C7FC48 141E121EA25AA25AA348143CA3153815785A15F0A2EC01E015C01403D8F01E13803970610700EB 810E00385B001C5B000E13E039078380403801FD00EA0001EC808014811500EB0387EB01FEA25C EB00F01F2D79A225>I<90B512C090380F0070153C151C011E130EA2150FA249131EA3153C4913 381570EC01E0EC07809038FFFC00EBF00E80EC0380D801E013C0A43903C00780A43907800F0015 01A2EC0702120F39FFF8038CC812F020237DA124>I<903801F02090380E0C4090381802C0EB30 01136001E0138013C01201A200031400A291C7FCA27FEA01F813FF6C13E06D7EEB1FF8EB03FCEB 007C143C80A30020131CA3141800601338143000705B5C38C80180D8C607C7FCEA81FC1B247DA2 1B>I<001FB512F8391E03C03800181418123038200780A200401410A2EB0F001280A200001400 131EA45BA45BA45BA4485AA41203B5FC1D2277A123>I<393FFE03FF3903C00078156015204848 1340A448C71280A4001EEB0100A4481302A4485BA400705B12F05C12705C5C123038380180D818 02C7FCEA0E0CEA03F0202377A124>I<3BFFF03FF80FF83B1F0007C003C0001E91388001801700 1602140F5E001F13175E6C13275E144702C75B1487D901075B16C001025C0381C7FC130415C213 08EC03C4131015C8132015D0134001C013E0138001005B5D120E92C8FC120C14022D2376A131> 87 D97 DI<137EEA01C138030180EA0703EA0E07121C003CC7FC12381278A35AA45B12701302 EA300CEA1830EA0FC011157B9416>I<143CEB03F8EB0038A31470A414E0A4EB01C013F9EA0185 EA0705380E0380A2121C123C383807001278A3EAF00EA31410EB1C201270133C38305C40138C38 0F078016237BA219>I<13F8EA0384EA0E02121C123C1238EA7804EAF018EAFFE0EAF000A25AA4 1302A2EA6004EA7018EA3060EA0F800F157A9416>I<143E144714CFEB018F1486EB0380A3EB07 00A5130EEBFFF0EB0E00A35BA55BA55BA55BA45B1201A2EA718012F100F3C7FC1262123C182D82 A20F>II<13F0EA 0FE01200A3485AA4485AA448C7FC131FEB2180EBC0C0380F00E0A2120EA2381C01C0A438380380 A3EB070400701308130E1410130600E01320386003C016237DA219>I<13C0EA01E013C0A2C7FC A8121C12231243A25AA3120EA25AA35AA21340EA7080A3EA71001232121C0B217BA00F>I<13F0 EA0FE01200A3485AA4485AA448C7FCEB01E0EB0210EB0C70380E10F0A2EB2060EB4000EA1D8000 1EC7FCEA1FC0EA1C70487EA27F142038703840A3EB188012E038600F0014237DA216>107 DI<391C0F80F8392610C10C39476066063987807807A2EB0070A2000EEBE00EA44848485A A3ED38202638038013401570168015303A7007003100D83003131E23157B9428>II<137EEA01C338038180380701C0120E001C13E0123C12381278A338 F003C0A21480130700701300130E130CEA3018EA1870EA07C013157B9419>I<3801C1F0380262 183804741C3808780CEB700EA2141EEA00E0A43801C03CA3147838038070A2EBC0E0EBC1C03807 2380EB1E0090C7FCA2120EA45AA3EAFFC0171F7F9419>III<13FCEA018338020080EA0401EA0C03140090C7FC120F13F0EA07FC6C7EEA003E130F 7F1270EAF006A2EAE004EA4008EA2030EA1FC011157D9414>I<13C01201A4EA0380A4EA0700EA FFF8EA0700A2120EA45AA45AA31310EA7020A213401380EA3100121E0D1F7C9E10>I<001E1360 002313E0EA4380EB81C01283EA8701A238070380120EA3381C0700A31408EB0E101218121CEB1E 20EA0C263807C3C015157B941A>I<381E0380382307C0EA43871383EA8381EA8700A200071380 120EA3381C0100A31302A25B5BA2EA0C30EA03C012157B9416>I<001EEB60E00023EBE1F0EA43 80EB81C000831470D887011330A23907038020120EA3391C070040A31580A2EC0100130F000C13 023806138C3803E0F01C157B9420>I<3803C1E0380462103808347038103CF0EA203814601400 C65AA45BA314203861C04012F1148038E2C100EA4462EA383C14157D9416>I<001E1330002313 70EA438014E01283EA8700A2380701C0120EA3381C0380A4EB0700A35BEA0C3EEA03CEEA000EA2 5B1260EAF0381330485AEA80C0EA4380003EC7FC141F7B9418>I<3801E0203803F0603807F8C0 38041F80380801001302C65A5B5B5B5B5B48C7FC120248138038080100485AEA3F06EA61FCEA40 F8EA807013157D9414>I E /Fj 7 62 df48 D<12035AB4FC1207B1EA7FF00C157E9412 >III<13 30A2137013F012011370120212041208121812101220124012C0EAFFFEEA0070A5EA03FE0F157F 9412>II61 D E /Fk 29 120 df<127012F8A3127005057C840E>58 D<127012F812FCA212741204A41208A2 1210A212201240060F7C840E>I<14801301A2EB0300A31306A35BA35BA35BA35BA35BA3485AA4 48C7FCA31206A35AA35AA35AA35AA35AA311317DA418>61 D<811401811403A21407140BA21413 143314231443811481130114011302A21304130C1308131090381FFFF0EB2000136013405BA248 C7FC5A12024880120C121E3AFF800FFF8021237EA225>65 D<90387FFFF8903807800FED0780ED 03C090380F0001A216E0A2011E14C01503A2ED078049EB0F00151E5DEC01F090387FFFE0903878 007881815B150E150FA24848131EA35D485A5D5DEC01C00007EB0F80B500FCC7FC23227EA125> I<017FB512C090380780031500A249C7FCA21680A2131EA2158016004948C7FCA25C5CEB7FFEEB 7806A3EBF004A25DEC0002485AA25D150C4848130815185D15700007EB03F0B65A22227EA124> 69 D<90387FFFF0903807801C150F8190390F000380A4011E1307A3ED0F005B151E5D15704948 5AD97FFFC7FC0178C8FCA25BA4485AA4485AA41207EAFFFC21227EA11F>80 D<90387FFFE0903807803C150E81D90F001380150316C0A2011EEB0780A3ED0F0049131E5D1570 EC01C0D97FFEC7FC9038780780EC01C081496C7EA44848485AA44848485A1640A2020113801207 3AFFFC00E300C8123C22237EA125>82 D<903803F01090380E0C20903818026090382001E0EB40 0001C013C05B1201A200031480A21500A27FEA01F013FE3800FFE06D7EEB1FF8EB01FCEB003C14 1C80A30020130CA3140800601318141000705B5C00C85BD8C603C7FCEA81FC1C247DA21E>I<00 1FB512FE391E01E00E001814061230382003C0A200401404A2EB07801280A20000140049C7FCA4 131EA45BA45BA45BA41201B512C01F227EA11D>I<3A3FFE01FF803A03C0003C00153015104848 5BA448C75AA4001E5CA44849C7FCA4481302A400705B12F05C12705C5C6C5B5C6C48C8FCEA0606 EA01F821237DA121>I<3BFFF03FFC03FF3B1F8007E000786C486C481360A217401780A20207EB 0100A2020B1302A202135B02235BA202435B018013E0000701815BA2D981015B018314C001825C 018401E1C7FCA2018813E2A2019013E4A201A013E801C013F0A201805B120390C75AA200025C30 237DA12E>87 D97 DI<133FEBE080380380C0EA0701EA0E03121C003CC7FCA25AA35A A400701340A23830018038380200EA1C1CEA07E012157E9415>I<137CEA0382EA0701120E121C 1238EA7802EA7004EAFFF800F0C7FCA25AA41480A238700300EA3004EA1838EA0FC011157D9417 >101 D<141EEC638014C71301ECC30014801303A449C7FCA4EBFFF8010EC7FCA65BA55BA55BA4 136013E0A35B1270EAF18090C8FC1262123C192D7EA218>I<13E0A21201EA00C01300A9121E12 23EA4380A21283A2EA87001207120EA35AA25A132013401270A2EA3080EA3100121E0B227EA111 >105 D<14E01301A2EB00C01400A9131E1323EB43801383EA0103A338000700A4130EA45BA45B A45BA3EA70E0EAF0C0EAF1800063C7FC123E132C81A114>I<13F0EA0FE01200A3485AA4485AA4 48C7FC14F0EB0308EB0438380E08781310EB2030EB4000485A001FC7FC13C0EA1C70487EA27F14 1038703820A3EB184038E00C803860070015237DA219>II<393C07E01F3A46183061 803A47201880C03A87401D00E0EB801E141C1300000E90383801C0A4489038700380A2ED070016 044801E01308150EA2ED0610267001C01320D83000EB03C026157E942B>I<383C07C038461860 384720303887403813801300A2000E1370A44813E0A2EB01C014C1003813C2EB0382A2EB018400 701388383000F018157E941D>I<133EEBC180380380C0380700E0120E4813F0123CA25AA338F0 01E0A214C0130300701380EB07001306EA381C6C5AEA07E014157E9417>I<3803C0F03804631C EB740EEA0878EB7007A2140FEA00E0A43801C01EA3143C38038038A2EBC07014E038072180EB1E 0090C7FCA2120EA45AA3EAFFC0181F819418>I114 D<137E138138030080EA0201EA0603140090C7FC 120713F8EA03FE6C7EEA003FEB07801303127000F01300A2EAE002EA4004EA3018EA0FE011157E 9417>I<136013E0A4EA01C0A4EA0380EAFFFCEA0380A2EA0700A4120EA45AA31308EA3810A213 20121813C0EA07000E1F7F9E12>I<001EEB18180023EB383CD84380133EEC701E0083140E1506 EA87000007EBE004120EA3391C01C008A31510A2152001031340D80C0413C0390708E1003801F0 3E1F157E9423>119 D E /Fl 43 121 df12 D45 D49 DII<157815F8140114031407A2140F141F143F147F147714F7EB01E7EB03 C7EB07871407130F131E133C1378137013F0EA01E0EA03C0EA0780EA0F00A2121E5A5A5AB712F0 A3C7380FF800A9010FB512F0A3242E7EAD29>I<000C1438390FC003F890B5FC15F015E015C015 80ECFE005C14F091C7FC90C8FCA7EB0FF8EB7FFF90B512C09038F01FE09038800FF090380007F8 000E14FCC7120315FEA215FFA2121E123FEA7F8012FF13C015FE1380A2397F0007FC007C14F800 3CEB0FF06CEB1FE0390FC07FC06CB512800001EBFE0038003FE0202E7CAD29>II<1238123E003FB612C0A316804815005D5D5D5D387800010070 495A4A5A00F0495A4891C7FC141E143EC75A5CA2495AA213035C1307A2130FA2495AA3133FA513 7FA86D5AA2010EC8FC22307BAF29>I<157CA215FEA34A7EA24A7FA24A7FA34A7F157F021F7FEC 1E3FA2023E7FEC3C1F027C7FEC780FA202F87FECF0070101804A7EA20103814A7E0107814A7EA2 49B67EA24981011EC7123FA24981161F017C810178140FA2496E7EA2000182486C80B5D8C001B5 12FEA337317DB03E>65 DI<913A03FF8001 80023FEBF00349B5EAFC0F01079038007F1FD91FF8EB0FBFD93FE0EB03FFD9FF8013004890C812 7F4848153F485A171F485A001F160F5B003F1607A2127F5B94C7FCA212FFA9127FA36DED038012 3FA2121F6D1507000F17006C7E170E6C6C151E6C6C5D6C6D5CD93FE05CD91FF8EB03E0D907FFEB 3F800101D9FFFEC7FCD9003F13F80203138031317BB03C>II II73 D77 D80 D82 D<90391FF0018090B51203000314C73907F00FFF380F800148C7127F48141F 003E140F127E150712FEA215037EA26D90C7FC13E0EA7FFCEBFFE014FE6CEBFFC06C14F0816C80 0003806C806C6C1480131F010014C01407020013E0153FA2151F126000E0140FA316C07EA26CEC 1F807EB4EC3F0001C0137E9038FC01FC00F1B55AD8E03F13E0D8C00790C7FC23317BB02E>I85 D87 D97 D99 DII<14FF 010713C0011F13E090383FC7F090387F0FF813FE120113FC0003EB07F0EC03E0EC01C091C7FCA7 B512F8A3D803FCC7FCB3A8387FFFF0A31D327EB119>I<90391FF007E09039FFFE3FF0489038FF 7FF83907F83FF1390FE00FE1EDE0F03A1FC007F0601600003F80A5001F5CA26C6C485AA23907F8 3FC090B5C7FC00065B380E1FF090C9FC121EA2121F7F90B512C06C14F815FE6C806C15804815C0 001F15E048C7127F007EEC0FF04814071503A4007EEC07E06CEC0FC0D81FC0EB3F803A0FF801FF 006CB55AC614F0011F1380252F7E9F29>III107 D I<2703F007F8EB0FF000FFD93FFFEB7FFE4A6DB5FC913BF03FC1E07F803D0FF1C01FE3803FC03C 07F3000FE6001F01F602FC14E013FE495CA2495CB3B500C1B50083B5FCA340207D9F45>I<3903 F007F800FFEB3FFF4A7F9138F03FC03A0FF1C01FE03907F3000F01F68013FE5BA25BB3B500C1B5 1280A329207D9F2E>II<3901F80FF000FFEB7FFE 01F9B512809039FFE07FC0000F9038001FE06C48EB0FF001F8EB07F816FC150316FEA2150116FF A816FE1503A216FC15076D14F86DEB0FF06DEB1FE09039FBE07FC001F9B512009038F87FFEEC1F E091C8FCABB512C0A3282E7E9F2E>I<3803F03F00FFEB7FC09038F1FFE09038F3C7F0000FEB8F F83807F70F13F613FE9038FC07F0EC03E0EC0080491300B2B512E0A31D207E9F22>114 DI<1338A51378A313F8A2120112031207121FB5 12FEA33807F800B01407A70003130E13FC3801FE1C3800FFF8EB7FF0EB0FE0182E7EAD20>IIII< B5EBFFFCA3D807F8EB1F806C6CEB1E006C6C5B6C6C5B6E5A90387FC1E0133F90381FE3C090380F F7806DB4C7FC5C130313016D7E497F81497F9038079FF0EB0F0F90381E07F8496C7E496C7E01F0 7F48486C1380ED7FC00007143F3AFFF801FFFEA327207E9F2C>I E /Fm 84 125 df<90381F83E09038706E309038C07C78380180F8000313F03907007000A9B612C03907 007000B21478397FE3FF801D2380A21C>11 DII<90380FC07F90397031C080 9039E00B00402601801E13E00003EB3E013807003C91381C00C01600A7B712E03907001C011500 B23A7FF1FFCFFE272380A229>I34 D37 D<127012F812FCA212741204A412 08A21210A212201240060F7CA20E>39 D<132013401380EA01005A12061204120CA25AA25AA312 701260A312E0AE1260A312701230A37EA27EA2120412067E7EEA0080134013200B327CA413>I< 7E12407E7E12187E12041206A27EA2EA0180A313C01200A313E0AE13C0A312011380A3EA0300A2 1206A21204120C5A12105A5A5A0B327DA413>I<7FA538C08180EAE08338388E00EA0C98EA03E0 6C5A487EEA0C98EA388E38E08380EAC08138008000A511157DA418>I<497EB0B612FEA2390001 8000B01F227D9C26>I<127012F812FCA212741204A41208A21210A212201240060F7C840E>II<127012F8A3127005057C840E>I<14801301A2EB0300A31306A35BA35BA3 5BA35BA35BA3485AA448C7FCA31206A35AA35AA35AA35AA35AA311317DA418>II<13801203120F12F31203B3A9EA07C0EAFFFE0F217CA018>III<13021306130EA2131EA2132E134EA2138EA2EA010E1202A212 04A212081210A21220A212401280B512F838000E00A7131F3801FFF015217FA018>I<00101380 381E0700EA1FFF5B13F8EA13E00010C7FCA613F8EA130EEA1407381803801210380001C0A214E0 A4127012F0A200E013C01280EA4003148038200700EA1006EA0C1CEA03F013227EA018>I<137E EA01C138030080380601C0EA0E03121C381801800038C7FCA212781270A2EAF0F8EAF30CEAF406 7F00F81380EB01C012F014E0A51270A3003813C0A238180380001C1300EA0C06EA070CEA01F013 227EA018>I<12401260387FFFE014C0A23840008038C0010012801302A2485A5BA25B13301320 1360A313E05BA21201A41203A86C5A13237DA118>III<12 7012F8A312701200AB127012F8A3127005157C940E>I<127012F8A312701200AB127012F8A312 781208A41210A312201240A2051F7C940E>I61 D63 D<497EA3497EA3EB05E0A2EB0DF01308 A2497E1478A2497EA3497EA3497EA290B5FC3901000780A24814C000021303A24814E01401A200 0CEB00F0A2003EEB01F839FF800FFF20237EA225>65 DI<903807E0109038381830EBE006 3901C0017039038000F048C7FC000E1470121E001C1430123CA2007C14101278A200F81400A812 781510127C123CA2001C1420121E000E14407E6C6C13803901C001003800E002EB381CEB07E01C 247DA223>IIII<90 3807F00890383C0C18EBE0023901C001B839038000F848C71278481438121E15185AA2007C1408 1278A200F81400A7EC1FFF0078EB00F81578127C123CA27EA27E7E6C6C13B86C7E3900E0031890 383C0C08903807F00020247DA226>I<39FFFC3FFF390FC003F039078001E0AE90B5FCEB8001AF 390FC003F039FFFC3FFF20227EA125>I I<3803FFF038001F007FB3A6127012F8A2130EEAF01EEA401C6C5AEA1870EA07C014237EA119> I<39FFFC03FF390FC000F86C48136015405D4AC7FC14025C5C5C5C5C5C1381EB83C0EB87E01389 EB88F01390EBA078EBC03C13808080A26E7E8114036E7EA26E7E81486C7F3AFFFC07FF8021227E A126>III<39FF8007FF3907C000F81570D805E01320EA04F0A21378137C133C7F131F7FEB0780A2 EB03C0EB01E0A2EB00F014F81478143C143E141E140FA2EC07A0EC03E0A21401A21400000E1460 121FD8FFE0132020227EA125>IIIII<3803F020380C0C60EA18023830 01E0EA70000060136012E0A21420A36C1300A21278127FEA3FF0EA1FFE6C7E0003138038003FC0 EB07E01301EB00F0A214707EA46C1360A26C13C07E38C8018038C60700EA81FC14247DA21B>I< 007FB512F839780780780060141800401408A300C0140C00801404A400001400B3A3497E0003B5 FC1E227EA123>I<39FFFC07FF390FC000F86C4813701520B3A5000314407FA2000114806C7E90 38600100EB3006EB1C08EB03F020237EA125>I<3BFFF03FFC03FE3B1F8007E000F86C486C4813 701720A26C6C6C6C1340A32703C002F01380A33B01E004780100A33A00F0083C02A39039F8183E 06903978101E04A2137C90393C200F08A390391E400790A390390F8003E0A36D486C5AA36D5C01 0213002F237FA132>87 D<397FF807FF3907E001F83903C000E06D5B00015C6C6C48C7FC6D5AEB 7802EB7C04EB3E0CEB1E08EB1F10EB0FB0EB07A014C06D7E130180497EEB0278EB047CEB0C3EEB 081EEB101F9038300F80EB200701407F9038C003E0EB8001D801007F4813004880391F8001FC3A FFE007FFC022227FA125>II<12FEA212C0B3B3A912FEA207317BA40E>91 DI< 12FEA21206B3B3A912FEA207317FA40E>I97 D<120E12FE121E120EAB131FEB61C0EB8060380F0030000E1338143C141C141EA7141C143C1438 000F1370380C8060EB41C038083F0017237FA21B>II<14E013 0F13011300ABEA01F8EA0704EA0C02EA1C01EA38001278127012F0A7127012781238EA1801EA0C 0238070CF03801F0FE17237EA21B>II<133C 13C6EA018F1203130FEA0700A9EAFFF8EA0700B21380EA7FF8102380A20F>I<14703801F19838 071E18EA0E0E381C0700A2003C1380A4001C1300A2EA0E0EEA0F1CEA19F00010C7FCA21218A2EA 1FFE380FFFC014E0383800F0006013300040131812C0A300601330A2003813E0380E03803803FE 0015217F9518>I<120E12FE121E120EABEB1F80EB60C0EB80E0380F0070A2120EAF38FFE7FF18 237FA21B>I<121C123EA3121CC7FCA8120E12FE121E120EB1EAFFC00A227FA10E>II< 120E12FE121E120EABEB03FCEB01F014C01480EB02005B5B5B133813F8EA0F1CEA0E1E130E7F14 80EB03C0130114E0EB00F014F838FFE3FE17237FA21A>I<120E12FE121E120EB3ADEAFFE00B23 7FA20E>I<390E1FC07F3AFE60E183803A1E807201C03A0F003C00E0A2000E1338AF3AFFE3FF8F FE27157F942A>I<380E1F8038FE60C0381E80E0380F0070A2120EAF38FFE7FF18157F941B>III<3801F82038070460EA0E02EA1C01003813E0EA7800A25AA712701278EA3801121CEA0C02EA 070CEA01F0C7FCA9EB0FFE171F7E941A>III<1202A41206A3120E 121E123EEAFFF8EA0E00AB1304A6EA07081203EA01F00E1F7F9E13>I<000E137038FE07F0EA1E 00000E1370AD14F0A238060170380382783800FC7F18157F941B>I<38FFC1FE381E0078000E13 301420A26C1340A238038080A33801C100A2EA00E2A31374A21338A3131017157F941A>I<39FF 8FF8FF391E01E03C001CEBC018120EECE010A239070260201470A239038430401438A23901C818 80141CA23900F00D00140FA2EB6006A320157F9423>I<38FF83FE381F01F0380E00C06C138038 0381001383EA01C2EA00E41378A21338133C134E138EEA0187EB0380380201C0000413E0EA0C00 383E01F038FF03FE17157F941A>I<38FFC1FE381E0078000E13301420A26C1340A238038080A3 3801C100A2EA00E2A31374A21338A31310A25BA35B12F05B12F10043C7FC123C171F7F941A>I< 383FFFC038380380EA300700201300EA600EEA401C133C1338C65A5B12015B38038040EA07005A 000E13C04813805AEA7801EA7007B5FC12157F9416>III E /Fn 51 121 df12 D<13181378EA01F812FFA21201B3A7387FFF E0A213207C9F1C>49 DI<13FE3807FFC0380F07E0381E03F0123FEB81F8A3EA1F 0314F0120014E0EB07C0EB1F803801FE007F380007C0EB01F014F8EB00FCA2003C13FE127EB4FC A314FCEA7E01007813F8381E07F0380FFFC03801FE0017207E9F1C>I<14E013011303A2130713 0F131FA21337137713E7EA01C71387EA03071207120E120C12181238127012E0B512FEA2380007 E0A7EBFFFEA217207E9F1C>I<00101320381E01E0381FFFC0148014005B13F8EA1BC00018C7FC A4EA19FCEA1FFF381E0FC0381807E01303000013F0A214F8A21238127C12FEA200FC13F0A23870 07E0003013C0381C1F80380FFF00EA03F815207D9F1C>II<1260 1278387FFFFEA214FC14F8A214F038E0006014C038C00180EB0300A2EA00065B131C1318133813 78A25BA31201A31203A76C5A17227DA11C>I57 D<1470A214F8A3497EA2497EA3EB06FF80010E7FEB0C3FA201187F141F01387FEB300FA201607F 140701E07F90B5FCA239018001FCA200038090C7FCA20006147FA23AFFE00FFFF8A225227EA12A >65 DII IIIIII76 DIIII82 D<3801FC043807FF8C381F03FC383C 007C007C133C0078131CA200F8130CA27E1400B4FC13E06CB4FC14C06C13F06C13F86C13FC0003 13FEEA003FEB03FFEB007F143FA200C0131FA36C131EA26C133C12FCB413F838C7FFE000801380 18227DA11F>I<007FB61280A2397E03F80F00781407007014030060140100E015C0A200C01400 A400001500B3A20003B512F8A222227EA127>II87 D89 D97 DI II<13FE3807FF80380F87C0381E01E0 003E13F0EA7C0014F812FCA2B5FCA200FCC7FCA3127CA2127E003E13186C1330380FC0703803FF C0C6130015167E951A>II<3803FC1E380FFF7F381F0F8F383E07CF383C03C0007C13 E0A5003C13C0EA3E07381F0F80EBFF00EA13FC0030C7FCA21238383FFF806C13F06C13F84813FC EA380048133E00F0131EA40078133C007C137C383F01F8380FFFE00001130018217E951C>II<121C 123E127FA3123E121CC7FCA7B4FCA2121FB2EAFFE0A20B247EA310>I107 DI<3AFF07F007F090391F FC1FFC3A1F303E303E01401340496C487EA201001300AE3BFFE0FFE0FFE0A22B167E9530>I<38 FF07E0EB1FF8381F307CEB403CEB803EA21300AE39FFE1FFC0A21A167E951F>I<13FE3807FFC0 380F83E0381E00F0003E13F848137CA300FC137EA7007C137CA26C13F8381F01F0380F83E03807 FFC03800FE0017167E951C>I<38FF0FE0EB3FF8381FF07CEB803E497E1580A2EC0FC0A8EC1F80 A29038803F00EBC03EEBE0FCEB3FF8EB0FC090C8FCA8EAFFE0A21A207E951F>I114 DI<487EA41203A21207A2120F123FB5FCA2EA0F80AB EB8180A5EB8300EA07C3EA03FEEA00F811207F9F16>I<38FF01FEA2381F003EAF147E14FE380F 81BE3907FF3FC0EA01FC1A167E951F>I<39FFE01FE0A2390F800600A2EBC00E0007130CEBE01C 00031318A26C6C5AA26C6C5AA2EB7CC0A2137F6D5AA26DC7FCA2130EA21B167F951E>I<3AFFE7 FF07F8A23A1F007800C0D80F80EB0180147CA23A07C07E030014DE01E05B0003EBDF06EBE18FD8 01F15B01F3138C9038FB079C000014D8EBFE03017E13F0A2EB7C01013C5BEB380001185B25167F 9528>I<39FFE07FC0A2390F801C006C6C5A6C6C5AEBF0606C6C5A3800F980137F6DC7FC7F8049 7E1337EB63E0EBC1F03801C0F848487E3807007E000E133E39FF80FFE0A21B167F951E>I E /Fo 16 122 df<1238127C12FEA3127C1238070774861F>46 D64 D97 DI II<13 7F3801FFC0000713E04813F0381F81F8383F0078003C133C127C0078131EA2B512FEA400F0C7FC A21278A2007C131E7E381F803EEBE07C380FFFF8000313F06C13E038003F80171A7D991F>I<13 30137813FCA21378133090C7FCA6EA7FFCA4EA003CB2387FFFFCB512FEA26C13FC17267CA51F> 105 D<1303EB0780EB0FC0A2EB0780EB030090C7FCA6380FFFC0A4EA0003B3AA38200780EA700F 38F81F00EAFFFE6C5A5BEA1FE012337DA51F>I108 D<38FF87E0EB9FF0EBBFF8EBFFFC3807F83CEBE01E13C0A21380AE39FFFC7FF0A41C1A7F991F> 110 D<13FCEA03FF481380001F13E01387383E01F0387C00F800781378A248133CA76C137C0078 1378007C13F8A2383E01F0381F87E013FF000713806C1300EA00FC161A7C991F>I<38FFE07E90 38E1FF8001E713C013EF3801FF879038FE038049C7FC5B5BA35BABB512F0A41A1A7E991F>114 D<3803FC70380FFFF0123F5AEA7C03EAF801EAF000A3007C1300EA7FE0EA1FFF000713C0C613E0 EB03F0EB00F80070133C12F0A27E6C137C38FF01F8EBFFF0A200E713C038E1FE00161A7C991F> I<38FF83FEA43807801EAF143EA2EBC0FE6CB512F0A26C139F38007E1F1C1A7F991F>117 D<397FE07FE039FFF0FFF0A2397FE07FE03907000E0013800003131E141CEA01C0143C1438EA00 E0A26D5A1370A26D5AA36D5A131DA2EB0F80A2130791C7FCA3130EA35B1238EA7C3C13F8EA7FF0 5B6C5AEA0F801C277F991F>121 D E /Fp 52 122 df<127812FCA212FEA2127A1202A41204A3 12081210A21220124007127B8511>44 DI<127812FCA4127806067B85 11>I<137F3801C1C0380780F0380F0078000E1338487F003C131EA3487FA400F81480AF007814 00A46C131EA3001C131C6C5B000F13786C6C5A3801C1C0D8007FC7FC19297EA71E>48 D<13101370EA01F0120F12FE12F01200B3AD487E387FFFE0A213287BA71E>I<13FE3807FF8038 0E07E0381803F0382001F8130048137CA200F8137E7E143EA30078137EC7FC147CA214F8A2EB01 F014E0EB03C0EB07801400130E5B5B5B13605B38018002EA0300000613045A5A0010130C383FFF FC4813F8B5FCA217287DA71E>I<137F3803FFC0380701F0380C00F80010137C121C003E137E14 3EA2121E000C137EC7127CA214785C5C495A0107C7FC13FFEB01E06D7E147880143E80A21580A2 1230127812FCA215005A00405B143E00305B6C5B380F01F03803FFC0C66CC7FC19297EA71E>I< 1460A214E01301A21303A213051309A213111321A213411381A2EA01011202A212041208A21210 12301220124012C0B61280A2390001E000A8497E90387FFF80A219287EA71E>I<00181318001F 13F0EBFFE014C014801400EA11F80010C7FCA8137E38118380381600C0001813E000101370C712 78143CA3143EA3127012F8A3143C12800040137C14787E003013F0381801E0380E07C03807FF00 EA01FC17297DA71E>I I<12201238003FB51280A215005A3860000200405BA25C485B5CC7FC5C5CA249C7FC5B13021306 A25BA2131CA35BA31378A413F8A81370192A7DA81E>I<137F3801FFC0380381F038060078487F 001C131C00187F1238A3123CA2003E5B381F8018EBC038380FF0606C6C5A6CB45A6C90C7FC3800 7FC0497E38030FF8380603FC381C01FE3838007E00307F0070130FEC07805A1403A46C14000070 5B0078130600385B001E1338380F80F03803FFE0C66CC7FC19297EA71E>I<137F3801FFC03807 C1E0380F0070001E7F001C133C003C131C48131EA200F87FA41580A41278141F7EA2001C132F6C 134F6C13CF3803810FEA007E01001300A3141EA35C121C003E5B1470003C5B381801C0381C0780 D80FFEC7FCEA03F819297EA71E>I<1418A3143CA3147EA214FF149FA29038011F80140FA29038 0207C0A3496C7EA3496C7EA201187FEB1000A2497F157C90383FFFFC497F903840003EA2497FA3 48C7EA0F80A30002EC07C01207D81F80EB0FE0D8FFF090B5FCA2282A7EA92D>65 DI<02FF13100107EBE03090381FC07890393E000C7001F8EB06F0484813 03484813014848130048481470A248C812305A123E1610127EA2007C150012FCA9127C127E1610 123EA2123F6C15206C7E16606C6C14406C6C14806C6C13016C6CEB0300013E130E90381FC03890 3807FFE001001380242B7DA92B>II70 D<02FF13100107EBE03090381FC07890393E000C7001F8EB06F04848130348481301484813 0048481470A248C812305A123E1610127EA2007C150012FCA892B5FC127C007EEC03F01501123E A2123F7E6C7EA26C7E6C7E6C6C13026C7E013EEB0C7090391FC03830903907FFE0100100EB8000 282B7DA92F>I73 D<0003B5FCA2380007E01303B3AA1230127812FCA214C0EAF8070040138038200F00EA300EEA0C 3CEA03F0182A7DA81F>IIIII82 DI<007FB612F8A2397C007C00007015380060 151800401508A200C0150CA2481504A5C71400B3A614FE90B512FEA226297EA82B>II<3DFFFE03FFF803FFC0A23D0FE0003F80007E 006C486DC712186C7E6F6C1310A26C6C5E82A26C6C5EED13E0A26D16C0017CD933F05B1521137E 013E6E48C7FC1540A26D1502ED807CA2D90F805C913881003EA2D907C15C02C2131FA2D903E25C 02E4EB0F90A2D901F414A002F8EB07E0A201005D4A1303A202705C02601301A33A2A7FA83D>87 D97 DII<140F49B4FCA2EB001F 80AC133F3801C0CF3803802F380F001F121E001C7F123C127C1278A212F8A71278A27EA26C5B00 0E132F6CEB4F803901C18FF838007E0F1D2A7EA921>I<137E3803C380380700E0000E13F04813 70003C1378143848133CA212F8A2B512FC00F8C7FCA51278127C003C1304A26C1308000E13106C 13203801C0C038007F00161A7E991B>I<131FEB70C0EBE1E0EA01C31203EB81C0380780801400 A9EAFFFEA2EA0780B3A37FEAFFFEA2132A7FA912>III<120FEA1F80A4EA0F00C7FCA9EA0780 127FA2120F1207B3A2EAFFF8A20D297FA811>I107 DI<260781F813FC3AFF860E0307903A98070C0380D80FA0019013C00007903903D001E001 C013E0A2018013C0B13BFFFC7FFE3FFFA2301A7F9933>I<380783F838FF8C1CEB900E380FA007 0007148013C0A21380B139FFFCFFFCA21E1A7F9921>I<137F3801C1C038070070000E7F487F00 3C131EA2487FA200F81480A800781400A26C131EA26C5B000E13386C5B3801C1C0D8007FC7FC19 1A7E991E>I<380783F038FF8C1CEBB00F3907A007809038C003C0018013E0140115F0A2EC00F8 A715F01401A215E0EC03C013C0EC07809038A00E00EB983CEB87E00180C7FCAAEAFFFCA21D267F 9921>I<380787C038FF98E0EB91F0EA0FA1EA07C1EBC0E014005BB07FEAFFFEA2141A7F9917> 114 D<3807F840381C06C0EA3001EA6000A200E01340A27E6C1300127EEA7FF0EA3FFEEA0FFF00 03138038003FC01307388001E0A2EAC000A36C13C0EAF00100F8138038C40700EA83F8131A7E99 18>I<7FA41201A31203A21207120F381FFF80B5FC38078000AD1440A73803C08012013800E100 133E12257FA417>I<390780078000FF13FFA2000F130F00071307B0140FA2000313173901C027 C03900E047FCEB3F871E1A7F9921>I<39FFF00FF8A2390F8003C000071480EC01003803C002A2 13E000015BA26C6C5AA2EBF818EB7810A26D5AA2EB3E60EB1E40A26D5AA26DC7FCA313021D1A7F 9920>I<3AFFF1FFC1FFA23A0F803E0078D9001E1330D807801420A2141F6C6C48134014271580 2601E0671380144315C03A00F081C100A201F813E390387900E2A2017D13F66D1374A2011E1378 011C1338A2010C133001081310281A7F992B>I<39FFF00FF8A2390F8003C000071480EC010038 03C002A213E000015BA26C6C5AA2EBF818EB7810A26D5AA2EB3E60EB1E40A26D5AA26DC7FCA313 02A25BA35B1270EAF810A25B485AEA6080001FC8FC1D267F9920>121 D E /Fq 25 122 df45 D<150C151EA3153FA34B7EA34B7EA39138019FE0 A202037F150FA202077FEC0607A2020C7F1503A202187F1501A24A6C7EA34A6D7EA214E04A6D7E A20101814A131FA201038191B6FCA249810106C71207A249811603A2496E7EA3496E7EA2017082 0160157FA201F082EA03F8D80FFC4A487EB500C0013FEBFFC0A33A3D7DBC41>65 DI69 D80 D85 D97 D99 DII<143F903801FFC0903803E0E090380781 F090380F83F8EB1F07133E137EEC03F090387C01E001FCC7FCAEB512FCA3D800FCC7FCB3AC487E 387FFFFCA31D3D7FBC1A>I<903907F001F890393FFE0FFC90397C1F1E3E9038F007F03A01E003 E01C2603C00113080007ECF000000F80EB8000001F80A7000F5CEBC00100075C00035C6C6C485A 6D485A26037C1FC7FC38073FFE380607F090C9FC120EA3120FA2EA07C090B512C06C14FC6C14FF 6C1580000315C03A0780003FE0001FC7EA07F0003EEC01F8003C1400127C0078157C12F8A5007C 15F8A26CEC01F06CEC03E06C6CEB07C0D803E0EB1F00D801FC13FE39003FFFF00107138027397E A52B>III108 D<2701F803F8EB03F800FFD91FFFEB1FFF91 3B3C0F803C0F80913BE007C0E007C03D07F9C003E1C003E02601FB00D9F3007F0301140101FE02 FE80A2495CA2495CB3A5486C496C497EB500F0B500F0B512F0A344267EA549>I<3901F807F800 FFEB1FFEEC781F9138E00F803A07F98007C02601FB007F150301FE805BA35BB3A5486C497EB500 F1B512E0A32B267EA530>II<3901F80FF000FFEB3FFEECF01F9039F9C007C03A03FB0003E0D801 FE6D7E496D7E8249147EA2167F821780A2161F17C0A91780163FA217005E167E5E7F4B5A6D495A 01FB495A9039F9C00FC09026F8F03FC7FCEC3FFCEC0FE091C9FCAC487EB512F0A32A377EA530> I<3903F00F8000FFEB3FE0EC70F0ECC1F83807F1833801F30313F6EC01F0EC004001FC1300A45B B3A3487EB512F8A31D267EA522>114 D<90387F81803803FFE3380F807F381E001F00381307A2 481303A200F01301A37EA200FE90C7FCEA7F8013FC383FFFC06C13F06C13FC00037F6C7FD8001F 13801300EC1FC00040130F00C0EB07E014036C1301A47E15C06C13036C1480EC070000F7130E38 E3C03C38C0FFF8EB3FC01B287DA622>I<1318A51338A41378A213F8A2120112031207001FB5FC B6FCA2D801F8C7FCB2EC0180A91200EC030013FC137CEB3E066D5AEB0FF8EB03F019367EB421> III121 D E end %%EndProlog %%BeginSetup %%Feature: *Resolution 300 TeXDict begin %%EndSetup %%Page: 0 1 bop 133 776 a Fq(A)29 b(P)n(arallelism-Based)h(Analytic)h(Approac)n(h)c(to) 131 880 y(P)n(erformance)g(Ev)-5 b(aluation)31 b(Using)e(Application)757 984 y(Programs)302 1244 y Fp(Da)n(vid)19 b(K.)h(Bradley)379 b(John)20 b(L.)g(Larson)203 1319 y Fo(bradley@)q(csr)q(d.u)q(iu)q(c.e)q(du) 154 b(jlarson@csr)q(d.u)q(iu)q(c.e)q(du)217 1543 y Fp(Cen)n(ter)21 b(for)f(Sup)r(ercomputing)f(Researc)n(h)g(and)h(Dev)n(elopmen)n(t)295 1618 y(465)f(Computer)h(and)g(Systems)e(Researc)n(h)h(Lab)r(oratory)705 1692 y(1308)h(W.)f(Main)h(St.)638 1767 y(Urbana,)g(IL)g(61801-2307)293 1917 y(This)g(w)n(ork)g(w)n(as)g(supp)r(orted)g(in)g(part)h(b)n(y)f(the)g (National)182 1991 y(Science)e(F)-5 b(oundation)20 b(under)h(Gran)n(t)g(No.) 26 b(NSF)19 b(ASC)i(89-02829.)767 2133 y(April)f(1,)f(1993)p eop %%Page: 1 2 bop eop %%Page: 1 3 bop 830 768 a Fn(Abstract)73 883 y Fm(In)11 b(this)g(pap)q(er)g(w)o(e)g (presen)o(t)f(a)i(brief)e(o)o(v)o(erview)f(of)i(p)q(erformance)f(ev)m (aluation)h(and)h(b)q(enc)o(hmarking.)0 943 y(W)l(e)k(demonstrate)f(that)i (traditional)f(p)q(erformance)e(measuremen)o(ts)g(recorded)h(in)h(these)g (activities)0 1003 y(are)i(really)g(a)h(direct)e(measuremen)o(t)e(of)k(the)f (parallelism)e(in)i(soft)o(w)o(are)h(and)g(hardw)o(are.)28 b(A)18 b(frame-)0 1063 y(w)o(ork)c(called)e(the)i(P)o(ath)g(to)g(P)o (erformance)e(is)h(dev)o(elop)q(ed)g(that)h(iden)o(ti\014es)e(the)h(agen)o (ts)h(and)h(activities)0 1123 y(that)j(c)o(hange)g(the)g(parallelism)e(as)i (it)g(mo)o(v)o(es)d(from)i(problem)g(to)h(solution.)27 b(W)l(e)17 b(sho)o(w)i(where)e(v)m(ar-)0 1183 y(ious)23 b(curren)o(t)e(application)h(b)q (enc)o(hmarks)f(apply)h(prob)q(es)h(on)g(the)f(P)o(ath)h(and)g(whic)o(h)e (agen)o(ts)i(are)0 1244 y(b)q(eing)c(measured.)29 b(W)l(e)19 b(recast)g(the)g(traditional)g(time-based)f(p)q(erformance)g(measuremen)o(ts) e(in)o(to)0 1304 y(parallelism-based)j(p)q(erformance)g(measuremen)n(ts)f(to) j(sho)o(w)f(that)h(understanding)g(p)q(erformance)0 1364 y(implies)12 b(understanding)j(the)g(parallelism.)j(A)c(sim)o(ulation)f(to)q(ol)i(and)h (metho)q(dology)e(are)h(describ)q(ed)0 1424 y(for)e(measuring)e(and)i (comparing)f(the)g(executed)f(parallelism)g(on)i(a)f(single)g(CRA)l(Y)g(Y-MP) g(CPU.)g(W)l(e)0 1484 y(apply)17 b(our)g(metho)q(dology)g(to)g(sev)o(eral)f (of)i(the)e(P)o(erfect)g(Benc)o(hmarks)f(to)i(quan)o(tify)f(their)g(executed) 0 1545 y(parallelism)c(on)j(this)f(mac)o(hine.)k(Our)c(results)g(suggest)i (that)e(since)g(some)f(of)i(the)f(b)q(enc)o(hmarks)f(ha)o(v)o(e)0 1605 y(a)k(similar)d(mix)h(of)i(di\013eren)o(t)e(lev)o(els)g(of)i (parallelism)d(on)j(the)f(Y-MP)l(,)g(the)g(b)q(enc)o(hmarking)f(utilit)o(y)g (of)0 1665 y(these)k(programs)g(on)g(the)g(Y-MP)g(ma)o(y)e(b)q(e)i (questioned)g(b)q(ecause)g(they)g(exercise)e(the)i(mac)o(hine)d(in)0 1725 y(the)g(same)f(w)o(a)o(y)l(.)931 2828 y(i)p eop %%Page: 2 4 bop 924 2828 a Fm(ii)p eop %%Page: 3 5 bop 0 203 a Fl(Con)n(ten)n(ts)0 312 y Fn(1)45 b(In)n(tro)r(duction)1465 b(1)0 421 y(2)45 b(P)n(erformance)18 b(Ev)m(aluation)f(as)i(a)g(Scien)n (ti\014c)f(Discipline)572 b(1)0 530 y(3)45 b(Preserving)17 b(and)j(Exploiting)c(P)n(arallelism)841 b(3)73 590 y Fm(3.1)50 b(Soft)o(w)o(are)17 b(P)o(arallelism)43 b Fk(:)25 b(:)f(:)h(:)f(:)g(:)h(:)f (:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f (:)g(:)h(:)94 b Fm(3)73 651 y(3.2)50 b(Hardw)o(are)17 b(P)o(arallelism)j Fk(:)25 b(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f (:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)94 b Fm(5)73 711 y(3.3)50 b(Realizing)15 b(P)o(arallelism)f(at)j(Run)o(time)40 b Fk(:)25 b(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f (:)h(:)f(:)g(:)h(:)94 b Fm(5)73 771 y(3.4)50 b(Benc)o(hmarking)14 b(Metho)q(dologies)j(and)g(the)f(P)o(ath)h Fk(:)24 b(:)h(:)f(:)g(:)h(:)f(:)h (:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)94 b Fm(5)73 831 y(3.5)50 b(The)17 b(P)o(arallelism)c(Matrix:)21 b(A)16 b(Measure)g(of)g(Executed)f(P)o (arallelism)40 b Fk(:)25 b(:)f(:)h(:)f(:)g(:)h(:)94 b Fm(7)73 891 y(3.6)50 b(Quan)o(tifying)16 b(Di\013erences)f(in)h(Executed)g(P)o (arallelism)32 b Fk(:)24 b(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:) 94 b Fm(8)0 1000 y Fn(4)45 b(Rederiv)m(ation)17 b(of)h(Basic)h(P)n(arallel)f (P)n(erformance)f(Metrics)511 b(9)73 1061 y Fm(4.1)50 b(P)o(erformance)21 b Fk(:)j(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f (:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)94 b Fm(9)73 1121 y(4.2)50 b(Execution)16 b(Rate)35 b Fk(:)24 b(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h (:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)69 b Fm(10)73 1181 y(4.3)50 b(P)o(eak)16 b(Execution)g(Rate)30 b Fk(:)25 b(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f (:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)69 b Fm(10)73 1241 y(4.4)50 b(Sp)q(eedup)32 b Fk(:)24 b(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h (:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h (:)f(:)g(:)h(:)69 b Fm(11)73 1301 y(4.5)50 b(E\016ciency)43 b Fk(:)25 b(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f (:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)69 b Fm(12)73 1362 y(4.6)50 b(Utilization)22 b Fk(:)j(:)f(:)h(:)f(:)h(:)f(:)h(:) f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h (:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)69 b Fm(12)73 1422 y(4.7)50 b(Amdahl's)15 b(La)o(w)20 b Fk(:)25 b(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f (:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f (:)g(:)h(:)69 b Fm(13)0 1531 y Fn(5)45 b(Using)19 b(P)n(arallelism)d(to)j(Ev) m(aluate)e(Application)h(Benc)n(hmarks)395 b(13)73 1591 y Fm(5.1)50 b(Generating)17 b(P)o(arallelism)c(Matrices)j(for)g(the)g(CRA)l(Y)g(Y-MP)23 b Fk(:)i(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)69 b Fm(14)73 1651 y(5.2)50 b(P)o(arallelism)14 b(Matrices)h(for)i(Selected)e(P)o(erfect)g (Benc)o(hmarks)29 b Fk(:)c(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)g(:)h(:)69 b Fm(15)73 1711 y(5.3)50 b(Di\013erences)16 b(in)g(Executed)f(P)o(arallelism) f(for)i(Selected)f(P)o(erfect)g(Benc)o(hmarks)k Fk(:)24 b(:)h(:)69 b Fm(16)73 1771 y(5.4)50 b(Prob)q(e)17 b(P)o(oin)o(ts)f(for)h(the)f(Baseline) f(P)o(erfect)g(Benc)o(hmarks)24 b Fk(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f(:)h(:)f (:)g(:)h(:)69 b Fm(18)0 1880 y Fn(6)45 b(Conclusions)19 b(and)g(F)-5 b(uture)19 b(W)-5 b(ork)1013 b(19)0 1989 y(7)45 b(Ac)n(kno)n(wledgmen)n(ts) 1306 b(19)917 2828 y Fm(iii)p eop %%Page: 4 6 bop 918 2828 a Fm(iv)p eop %%Page: 1 7 bop 0 203 a Fl(1)83 b(In)n(tro)r(duction)0 313 y Fm(Since)17 b(their)h(inception,)f(the)h(p)q(erformance)f(of)h(computer)f(systems)g(has)h (enjo)o(y)o(ed)f(unpreceden)o(ted)0 373 y(gro)o(wth.)k(F)l(or)14 b(decades)f(the)g(enabling)g(force)h(b)q(ehind)f(this)g(gro)o(wth)h(has)h(b)q (een)e(the)g(steady)h(and)g(rapid)0 433 y(impro)o(v)o(em)o(en)n(t)g(of)i(the) h(basic)f(circuit)f(tec)o(hnology)l(.)21 b(Computers)16 b(b)q(ecame)f(more)g (p)q(o)o(w)o(erful)h(b)q(ecause)0 493 y(the)d(basic)f(comp)q(onen)o(ts)h (used)g(in)f(their)g(construction)h(|)g(from)f(tub)q(es)h(to)g(transistors)h (to)g(successiv)o(e)0 554 y(generations)g(of)g(in)o(tegrated)e(circuit)g(tec) o(hnology)h(|)h(b)q(ecame)e(faster,)h(smaller,)f(more)g(e\016cien)o(t,)f(and) 0 614 y(less)19 b(exp)q(ensiv)o(e.)29 b(The)19 b(con)o(tributions)g(of)g(arc) o(hitectural)f(inno)o(v)m(ations,)i(although)g(n)o(umerous)e(and)0 674 y(signi\014can)o(t,)e(w)o(ere)f(dw)o(arfed.)73 734 y(F)l(or)j(w)o(ell)e (o)o(v)o(er)h(a)h(decade)f(prop)q(onen)o(ts)h(of)g(parallel)f(computer)f (systems)g(ha)o(v)o(e)h(predicted)g(that)0 794 y(circuit)f(tec)o(hnology)h(w) o(ould)g(so)q(on)i(reac)o(h)d(limits)f(imp)q(osed)h(b)o(y)h(the)g(la)o(ws)g (of)h(ph)o(ysics,)e(and)i(this)f(ap-)0 855 y(p)q(ears)k(to)g(b)q(e)f(happ)q (ening.)34 b(The)20 b(clo)q(c)o(k)g(rates)g(of)h(the)f(fastest)g(sup)q (ercomputers)g(and)h(the)f(fastest)0 915 y(micropro)q(cessors)15 b(no)o(w)h(di\013er)f(b)o(y)f(only)i(a)f(factor)h(of)g(three.)1106 897 y Fj(1)1146 915 y Fm(W)l(e)f(are)g(en)o(tering)g(a)g(new)h(era)f(in)g (com-)0 975 y(puter)j(arc)o(hitecture)f(in)i(whic)o(h)f(almost)f(all)h (signi\014can)o(t)h(p)q(erformance)e(impro)o(v)o(em)o(en)o(ts)e(will)j(o)q (ccur)0 1035 y(b)o(y)f(exploiting)g(increased)g Fi(p)n(ar)n(al)r(lelism)i Fm(rather)f(than)g(faster)g(basic)g(tec)o(hnology)l(.)26 b(Consequen)o(tly)17 b(if)0 1095 y(the)k(\014eld)f(of)h(P)o(erformance)f(Ev)m(aluation)h(is)g(to)g (pro)o(vide)f(meaningful)g(information)g(to)h(designers)0 1156 y(and)16 b(users)f(of)g(high-p)q(erformance)g(hardw)o(are,)g(new)g Fi(p)n(ar)n(al)r(lelism-b)n(ase)n(d)h Fm(measures)e(of)h(p)q(erformance)0 1216 y(m)o(ust)f(b)q(e)h(dev)o(elop)q(ed)f(to)h(complemen)n(t)d(the)j (traditional)g Fi(time-b)n(ase)n(d)h Fm(measures)e(suc)o(h)g(as)i(execution)0 1276 y(rate.)73 1336 y(This)j(pap)q(er)h(examines)d(P)o(erformance)g(Ev)m (aluation)i(from)f(the)h(viewp)q(oin)o(t)f(that)h(ev)o(en)f(though)0 1396 y(reducing)11 b(execution)f(time)f(is)j(the)f(ultimate)e(goal,)k (parallelism)c(is)i(the)g(essen)o(tial)f(c)o(haracteristic)g(nec-)0 1457 y(essary)15 b(for)h(ac)o(hieving)e(go)q(o)q(d)j(p)q(erformance.)j (Section)15 b(2)g(examines)e(P)o(erformance)h(Ev)m(aluation)i(as)g(a)0 1517 y(Scien)o(ti\014c)d(Discipline)f(with)j(sp)q(ecial)f(emphasis)f(on)i (the)f(role)g(of)h(careful)f(design)g(and)h(measuremen)o(t)0 1577 y(in)c(the)g(P)o(erformance)f(Ev)m(aluation)i(pro)q(cess.)21 b(Section)11 b(3)g(in)o(tro)q(duces)h(the)f(P)o(ath)h(to)g(P)o(erformance)d (and)0 1637 y(sho)o(ws)19 b(that)g(the)g(k)o(ey)e(to)i(ac)o(hieving)e(high)i (p)q(erformance)f(is)g(preserving)g(parallelism)e(across)k(eac)o(h)0 1697 y(stage)g(of)f(the)g(soft)o(w)o(are)g(dev)o(elopmen)o(t)d(pro)q(cess)k (and)g(b)o(y)e(pro)o(viding)h(hardw)o(are)g(that)h(can)f(exploit)0 1758 y(this)i(parallelism.)32 b(This)21 b(section)g(also)g(in)o(tro)q(duces)g (a)g(simple)e(example)f(of)j(a)h(parallelism-based)0 1818 y(metric,)15 b(the)j(P)o(arallelism)d(Matrix.)25 b(F)l(undamen)o(tal)17 b(time-based)f(measuremen)o(ts)f(of)j(p)q(erformance)0 1878 y(are)j(recast)f(as)i(parallelism-based)d(measuremen)n(ts)f(in)j(Section)f (4,)i(demonstrating)e(that)h(under-)0 1938 y(standing)h(p)q(erformance)f (implies)e(understanding)j(the)f(parallelism.)35 b(Section)21 b(5)h(describ)q(es)f(ho)o(w)0 1998 y(parallelism-based)13 b(p)q(erformance)h (metrics)f(migh)o(t)h(b)q(e)h(applied)f(to)i(the)f(design)g(of)g(b)q(enc)o (hmark)f(sets)0 2058 y(of)19 b(application)g(programs,)g(and)h(applies)f(suc) o(h)f(metrics)f(to)i(the)g(P)o(erfect)f(Benc)o(hmarks.)27 b(Our)19 b(re-)0 2119 y(sults)h(p)q(oin)o(t)f(to)h(some)e(p)q(ossible)i(shortcomings)f (in)g(curren)o(t)g(metho)q(dology)g(used)h(for)g(b)q(enc)o(hmark)0 2179 y(construction.)h(Finally)l(,)14 b(w)o(e)h(presen)o(t)g(our)h (conclusions)f(and)i(plans)e(for)h(future)f(w)o(ork)h(in)f(Section)g(6.)0 2345 y Fl(2)83 b(P)n(erformance)23 b(Ev)-5 b(aluation)25 b(as)f(a)g(Scien)n (ti\014c)e(Discipline)0 2455 y Fm(In)16 b(an)o(y)g(scien)o(ti\014c)e(or)i (engineering)f(discipline,)f(progress)j(is)f(ac)o(hiev)o(ed)e(b)o(y)i(ev)m (aluating)g(the)f(state)i(of)0 2515 y(the)e(art)g(to)g(\014nd)g(w)o(a)o(ys)g (in)f(whic)o(h)g(it)g(can)h(b)q(e)g(impro)o(v)o(ed.)j(The)d(e\016ciency)e (with)h(whic)o(h)g(this)h(progres-)p 0 2559 750 2 v 56 2589 a Fh(1)75 2604 y Fg(F)m(or)g(example,)f(2)i(nanoseconds)h(for)e(the)h(Cra)o (y-3)f(v)o(ersus)i(6.7)e(nanoseconds)i(\(150)e(Mhz\))h(for)g(the)g(DEC)f (Alpha)0 2654 y(mo)q(del)d(3000.)925 2828 y Fm(1)p eop %%Page: 2 8 bop 0 195 a Fm(sion)18 b(mo)o(v)o(es)d(forw)o(ard)j(dep)q(ends)g(in)f(part)h (on)f(the)h(accuracy)f(and)h(precision)e(of)i(the)f(metho)q(ds)g(used)0 255 y(to)e(ev)m(aluate)f(curren)o(t)g(practice.)20 b(Metho)q(ds)15 b(that)g(do)g(not)g(facilitate)e(clear)h(understanding)i(of)f(exist-)0 315 y(ing)g(problems)e(and)i(p)q(oten)o(tial)g(solutions)g(w)o(aste)f(ph)o (ysical,)g(\014nancial,)g(and)h(in)o(tellectual)e(resources.)0 376 y(Ev)m(aluation,)18 b(therefore,)f(is)h(a)g(cornerstone)g(of)g(sound)h (scien)o(ti\014c)d(and)j(engineering)e(practice.)24 b(Ac-)0 436 y(curate)16 b(and)h(meaningful)e(measuremen)n(t,)e(in)j(turn,)g(is)g(a)h (prerequisite)d(of)j(go)q(o)q(d)h(ev)m(aluation.)73 496 y(In)e(order)h(to)g (cast)f(P)o(erformance)f(Ev)m(aluation)i(as)g(scien)o(ti\014c)e(discipline,)f (one)j(m)o(ust)e(\014rst)h(de\014ne)0 556 y(\\p)q(erformance")23 b(and)i(\\ev)m(aluation.")44 b(The)24 b(p)q(erformance)f(of)h(a)g(computer)e (system)h(running)h(a)0 616 y(program)15 b(can)g(b)q(e)h(measured)e(b)o(y)g (sev)o(eral)g(criteria.)20 b(Of)15 b(ma)s(jor)f(imp)q(ortance)g(is)h(the)g (elapsed)g(time)e(a)0 677 y(computer)d(system)g(requires)g(to)i(solv)o(e)f(a) h(problem.)18 b(By)11 b(con)o(v)o(en)o(tion)f(this)h(time)f(is)h(usually)g (measured)0 737 y(b)o(y)g(its)h(in)o(v)o(erse,)e(the)h Fi(r)n(ate)h Fm(at)g(whic)o(h)f(solutions)h(are)g(obtained.)20 b(This)11 b(rate)h(is)g(called)e(the)i Fi(p)n(erformanc)n(e)0 797 y Fm(of)g(the)f (system.)19 b(Prerequisite)10 b(to)i(ev)m(aluating)f(\(i.e.)19 b(understanding)12 b(and)g(impro)o(ving\))e(p)q(erformance)0 857 y(is)21 b(the)g(abilit)o(y)f(to)h Fi(me)n(asur)n(e)g Fm(p)q(erformance.) 35 b(The)21 b(practice)f(of)h(measuring)g(the)g(p)q(erformance)f(of)0 917 y(computer)11 b(systems)h(is)h(commonly)d(called)h Fi(b)n(enchmarking)p Fm(,)k(and)e(programs)g(used)g(to)g(p)q(erform)f(these)0 978 y(measuremen)o(ts)h(are)k(called)e Fi(b)n(enchmarks)p Fm(.)794 960 y Fj(2)73 1038 y Fm(As)i(with)f(an)o(y)h(scien)o(ti\014c)e(discipline,)f (one)j(uses)g(the)f(basic)h(steps)g(of)g(the)f(Scien)o(ti\014c)f(Metho)q(d)i (to)0 1098 y(conduct)i(meaningful)e(exp)q(erimen)o(ts)f(and)j(comm)o(unic)o (ate)d(understandable)j(results.)29 b(The)18 b(\\disci-)0 1158 y(pline")d(of)g(P)o(erformance)f(Ev)m(aluation,)h(therefore,)g(consists)g(of) h(four)g(di\013eren)o(t,)e(in)o(terrelated)f(activ-)0 1218 y(ities,)i(striving)h(to)g(answ)o(er)h(basic)f(p)q(erformance-related)f (questions:)60 1322 y(1.)24 b(Design)f({)f(What)h(p)q(erformance)e(are)i(w)o (e)f(trying)g(to)h(measure)e(and)i(ho)o(w)f(do)h(w)o(e)f(measure)122 1383 y(it?)f(Activities)14 b(include)h(creating)h(prop)q(er)h(exp)q(erimen)o (ts)c(and)k(metho)q(dologies)f(for)g(testing)h(a)122 1443 y(p)q(erformance)e (h)o(yp)q(othesis,)60 1542 y(2.)24 b(Observ)m(ation)15 b({)g(What)g(happ)q (ened)h(in)e(the)h(exp)q(erimen)o(t?)j(Activities)12 b(include)i(executing)f (the)122 1602 y(exp)q(erimen)o(t)g(and)k(recording)f(measuremen)o(ts,)d(coun) o(ts,)j(traces,)g(etc.,)60 1701 y(3.)24 b(Analysis)12 b({)g(What)h(caused)g (the)f(b)q(eha)o(vior)g(observ)o(ed)g(during)h(the)f(p)q(erformance)f(exp)q (erimen)o(t?)122 1762 y(Activities)j(include)h(understanding)i(what)g(the)f (recorded)f(n)o(um)o(b)q(ers)g(mean.)60 1861 y(4.)24 b(Syn)o(thesis)13 b({)g(Giv)o(en)g(what)h(w)o(as)g(learned)f(from)f(the)h(exp)q(erimen)o(t,)e (ho)o(w)i(can)h(the)f(p)q(erformance)122 1921 y(b)q(e)18 b(impro)o(v)o(ed?)24 b(Activities)15 b(include)h(pro)o(viding)i(feedbac)o(k)e(to)j(hardw)o(are)f (and)g(soft)o(w)o(are)g(de-)122 1981 y(signers)e(regarding)h(new)f(opp)q (ortunities)h(for)g(higher)f(p)q(erformance.)0 2085 y(These)d(four)h(steps)g (are)g(rep)q(eated)f(un)o(til)f(the)i(desired)f(understanding)h(of)g(p)q (erformance)e(is)h(ac)o(hiev)o(ed.)73 2146 y(A)19 b(basic)h(to)q(ol)g(used)g (in)g(P)o(erformance)e(Ev)m(aluation)i(is)f(the)h(b)q(enc)o(hmark.)30 b(A)19 b(b)q(enc)o(hmark)f(is)i(a)0 2206 y(w)o(ell-de\014ned)d(sp)q (eci\014cation)i(of)g(an)g(exp)q(erimen)o(t)c(including)j(1\))h(a)g(problem)e (description)h(|)h(prose)0 2266 y(or)14 b(a)g(set)g(of)g(programs,)g(2\))g(a) g(metho)q(dology)f(or)h(set)g(of)g(rules)f(for)h(conducting)g(the)g(exp)q (erimen)o(t,)d(3\))j(a)0 2326 y(set)g(of)g(desired)f(measuremen)n(ts,)e(and)k (4\))f(a)g(v)o(eri\014cation)e(test)i(indicating)f(whether)g(the)h(exp)q (erimen)o(t)0 2386 y(succeeded)e(in)h(pro)q(ducing)h(a)g(correct)f(solution)g (to)h(the)f(problem.)19 b(Benc)o(hmarking)11 b(is)i(the)g(execution)0 2447 y(of)k(the)g(b)q(enc)o(hmark)e(on)j(the)f(mac)o(hine\(s\))e(of)i(in)o (terest)f(and)h(the)g(rep)q(orting)g(of)h(the)e(recorded)h(infor-)0 2507 y(mation.)i(Benc)o(hmarking)11 b(is)i(part)g(of)h(the)f(observ)m(ation)h (activit)o(y)d(in)i(P)o(erformance)e(Ev)m(aluation.)21 b(All)0 2567 y(to)q(o)c(often,)f(ho)o(w)o(ev)o(er,)f(b)q(enc)o(hmarking)f(is)i (mistak)o(en)f(for)h(P)o(erformance)f(Ev)m(aluation.)p 0 2608 750 2 v 56 2639 a Fh(2)75 2654 y Fg(The)21 b(distinction)f(b)q(et)o(w)o(een)h (b)q(enc)o(hmarking)f(and)g(p)q(erformance)g(ev)n(aluation)f(is)i(often)f(o)o (v)o(erlo)q(ok)o(ed)g(ev)o(en)h(b)o(y)0 2704 y(practitioners.)925 2828 y Fm(2)p eop %%Page: 3 9 bop 73 195 a Fm(It)18 b(can)h(b)q(e)f(argued)h(that)g(often)g(in)f(curren)o (t)f(practice)h(to)q(o)h(m)o(uc)o(h)e(emphasis)g(is)h(placed)g(on)h(ob-)0 255 y(serv)m(ation)c(\(step)g(3\))g(|)g(v)m(arious)g(w)o(ell-kno)o(wn)f(b)q (enc)o(hmark)f(programs)i(are)g(run)g(on)g(a)g(new)g(mac)o(hine)0 315 y(or)i(sim)o(ulated)e(on)j(a)f(prop)q(osed)h(mac)o(hine,)d(and)i(the)g (timing)e(results)i(are)g(rep)q(orted)g(in)g(v)o(oluminous)0 376 y(tables)d(and)h(graphs.)22 b(W)l(e)14 b(w)o(ould)g(lik)o(e)f(to)i(see)e (more)h(atten)o(tion)g(giv)o(en)f(to)i(the)f(analysis)g(and)h(syn)o(the-)0 436 y(sis)g(steps,)g(and)h(esp)q(ecially)d(to)i(the)g(design)g(of)g(b)q(enc)o (hmarks.)20 b(This)15 b(pap)q(er)g(addresses)h(the)f(design)g(of)0 496 y(b)q(enc)o(hmark)g(sets)h(b)o(y)g(prop)q(osing)i(a)e(parallelism-based)f (criterion)g(for)h(b)q(enc)o(hmark)f(comparison.)0 662 y Fl(3)83 b(Preserving)26 b(and)h(Exploiting)h(P)n(arallelism)73 772 y Fm(If)19 b(w)o(e)h(lo)q(ok)g(at)g(the)g(en)o(tire)e(computational)h(pro)q (cess)i(from)d(problem)h(de\014nition)g(to)h(program)0 832 y(execution,)15 b(w)o(e)h(\014nd)h(that)g(some)e(asp)q(ect)i(of)g(p)q (erformance)e(is)h(in\015uenced)g(b)o(y)g(eac)o(h)g(step.)22 b(Figure)16 b(1)0 892 y(depicts)f(what)i(w)o(e)e(call)g(the)h(P)o(ath)g(T)l (o)h(P)o(erformance.)i(The)d(left)f(side)g(of)h(the)g(\014gure)g(represen)o (ts)f(the)0 953 y(soft)o(w)o(are)h(dev)o(elopmen)o(t)e(pro)q(cess,)i(while)g (the)g(righ)o(t)g(sho)o(ws)h(the)f(hardw)o(are)h(dev)o(elopmen)o(t)c(pro)q (cess.)0 1013 y(W)l(e)i(call)f(eac)o(h)g(abstraction)i(a)f Fi(step)g Fm(in)g(the)f(P)o(ath.)22 b(The)14 b(P)o(ath)i(is)e(tra)o(v)o (ersed)g(through)i(the)f(activities)0 1073 y(of)24 b(a)f(team)f(of)h Fi(agents)i Fm(that)f(transform)e(the)h(soft)o(w)o(are)h(and)f(hardw)o(are)h (from)e(one)i(step)f(to)g(the)0 1133 y(next.)30 b(The)19 b(soft)o(w)o(are)g (and)h(hardw)o(are)g(meet)d(when)i(a)h(program)f(is)g(executed,)f(resulting,) h(from)g(a)0 1193 y(p)q(erformance)h(viewp)q(oin)o(t,)h(in)g(a)h(measure)e (of)h(program)g(execution)f(time.)34 b(Finally)l(,)21 b(a)h(metric)c(is)0 1253 y(applied)i(to)g(the)g(execution)f(time)f(to)j(get)f(a)g(measure)f(of)i (p)q(erformance.)1388 1235 y Fj(3)1439 1253 y Fm(With)f(the)g(tenet)g(that)0 1314 y(p)q(erformance)13 b(is)g(parallelism)f(\()p Ff(x)p Fm(4\),)i(w)o(e)f (see)g(that)i(three)e(conditions)h(m)o(ust)e(b)q(e)i(met)e(to)i(ac)o(hiev)o (e)e(high)0 1374 y(p)q(erformance.)19 b(First,)13 b(the)g(parallelism)d (presen)o(t)j(in)f(the)h(ph)o(ysical)f(problem)g(m)o(ust)f(b)q(e)i(preserv)o (ed,)f(as)0 1434 y(m)o(uc)o(h)j(as)i(p)q(ossible,)g(through)g(eac)o(h)g (phase)g(in)f(the)h(soft)o(w)o(are)g(dev)o(elopmen)o(t)c(pro)q(cess.)24 b(Second,)16 b(the)0 1494 y(hardw)o(are)j(m)o(ust)d(pro)o(vide)h(su\016cien)o (t)g(parallelism)f(to)i(realize)f(the)h(desired)f(p)q(erformance.)26 b(Third,)0 1554 y(the)16 b(soft)o(w)o(are)g(and)h(hardw)o(are)g(parallelism)d (m)o(ust)h(b)q(e)h(compatible.)73 1615 y(Eac)o(h)k(soft)o(w)o(are)g(and)h (hardw)o(are)f(agen)o(t)g(optimizes)e(t)o(w)o(o)i(of)g(the)g(three)f(v)m (ariables)h(that)g(deter-)0 1675 y(mine)e(p)q(erformance)g(|)h(clo)q(c)o(k)g (rate,)h(parallelism,)e(and)i(w)o(ork)f(\()p Ff(x)p Fm(4.1\))h(|)g(to)g(meet) d(its)i(ob)s(jectiv)o(e.)0 1735 y(Soft)o(w)o(are)j(agen)o(ts)g(m)o(ust)e (optimize)f(the)i(balance)h(b)q(et)o(w)o(een)f(w)o(ork)g(and)i(parallelism.) 35 b(Execution)0 1795 y(time)16 b(can)h(sometimes)e(b)q(e)j(decreased)f(b)o (y)g(increasing)h(the)f(amoun)o(t)g(of)h(w)o(ork,)f(if)g(the)h(resulting)f (in-)0 1855 y(crease)d(in)g(parallelism)e(is)j(su\016cien)o(tly)d(large.)21 b(Con)o(v)o(ersely)l(,)13 b(execution)g(time)f(can)j(b)q(e)g(shortened)g(b)o (y)0 1916 y(signi\014can)o(tly)h(decreasing)h(the)g(amoun)o(t)f(of)h(w)o (ork,)g(for)h(example)c(via)j(a)h(more)d(e\016cien)o(t)h(algorithm,)0 1976 y(ev)o(en)22 b(though)h(the)g(amoun)o(t)f(of)h(parallelism)d(ma)o(y)h(b) q(e)i(sligh)o(tly)f(smaller.)39 b(Similarly)-5 b(,)21 b(giv)o(en)h(the)0 2036 y(constrain)o(ts)e(sp)q(eci\014ed)f(b)o(y)g(the)g(system)f(requiremen)o (ts)f(\(cost,)j(p)q(o)o(w)o(er)f(consumption,)h(size,)f(etc.\),)0 2096 y(hardw)o(are)e(agen)o(ts)f(optimize)e(the)i(balance)g(b)q(et)o(w)o(een) g(parallelism)d(and)k(clo)q(c)o(k)f(rate.)0 2241 y Fe(3.1)70 b(Soft)n(w)n(are)23 b(P)n(arallelism)0 2333 y Fm(In)f(the)g(P)o(ath)g(T)l(o)h (P)o(erformance)d(the)i(soft)o(w)o(are)g(dev)o(elopmen)o(t)d(pro)q(cess)k (starts)g(with)f(a)g(ph)o(ysical)0 2393 y(phenomenon)13 b(and)i(ends)f(with)f (a)i(stream)d(of)j(mac)o(hine)c(instructions.)21 b(Man)o(y)13 b(ph)o(ysical)g(phenomena)0 2453 y(con)o(tain)18 b(signi\014can)o(t)g (inheren)o(t)f(parallelism;)g(extremely)e(high)k(p)q(erformance)e(can)h(b)q (e)h(obtained)g(if)p 0 2497 750 2 v 56 2528 a Fh(3)75 2543 y Fg(Often)13 b(the)g(wrong)g(metric)f(is)h(used,)g(giving)e(the)j(illusion)d (of)h(p)q(o)q(or)h(\(or)g(go)q(o)q(d\))f(p)q(erformance.)18 b(F)m(or)12 b(example,)f(the)0 2593 y(most)i(common)f(metric,)h(mega\015ops,) g(when)h(applied)g(to)g(a)g(pattern-matc)o(hing)g(program)e(will)h(giv)o(e)h (a)g(misleadingly)0 2642 y(small)f(v)n(alue,)i(implying)e(p)q(o)q(or)j(p)q (erformance.)23 b(In)15 b(this)h(case)h(a)e(di\013eren)o(t)h(metric)f(suc)o (h)i(as)f(comparisons)e(or)i(logical)0 2692 y(op)q(erations)e(w)o(ould)f(b)q (e)i(a)e(more)g(accurate)i(measure)f(of)f(\\w)o(ork")h(as)f(de\014ned)i(in)f Fd(x)p Fg(4.1.)925 2828 y Fm(3)p eop %%Page: 4 10 bop 319 556 a Fn(SOFTW)-6 b(ARE)277 616 y(P)h(ARALLELISM)1167 556 y(HARD)n(W)f(ARE)1136 616 y(P)h(ARALLELISM)245 737 y Fm(Ph)o(ysical)15 b(Phenomenon)474 797 y Ff(#)376 857 y Fi(\(Scientist\))474 917 y Ff(#)257 977 y Fm(Mathematical)f(Mo)q(del)474 1038 y Ff(#)270 1098 y Fi(\(Numeric)n(al)j(A)o(nalyst\))474 1158 y Ff(#)360 1218 y Fm(Algorithms)508 b(System)15 b(Requiremen)o(ts)474 1278 y Ff(#)273 1339 y Fi(\(Softwar)n(e)j(Engine)n(er\))474 1399 y Ff(#)1333 1278 y(#)1231 1339 y Fi(\(A)o(r)n(chite)n(ct\))1333 1399 y Ff(#)386 1459 y Fm(Program)549 b(System)15 b(Arc)o(hitecture)474 1519 y Ff(#)275 1579 y Fi(\(Compiling)j(System\))474 1640 y Ff(#)1333 1519 y(#)1122 1579 y Fi(\(Har)n(dwar)n(e)e(Engine)n(er\))1333 1640 y Ff(#)352 1700 y Fm(Instructions)423 b(Arc)o(hitecture)14 b(Implem)o(e)o(n)o(tation)454 1760 y Ff(&)809 b(.)544 1820 y(\000)-9 b(!)91 b Fi(\(Pr)n(o)n(gr)n(am)15 b(Exe)n(cution\))94 b Ff( )-8 b(\000)925 1880 y(#)767 1940 y Fm(Execution)16 b(Time)933 2001 y Ff(#)856 2061 y Fi(\(Metric\))933 2121 y Ff(#)712 2181 y Fn(PERF)n(ORMANCE)167 2334 y(Figure)i(1)p Fm(:)j(The)c(P)o(ath)f(T)l(o)h(P) o(erformance.)j(Agen)o(ts)15 b(are)i(sho)o(wn)g(in)f(paren)o(theses.)925 2828 y(4)p eop %%Page: 5 11 bop 0 195 a Fm(this)21 b(parallelism)f(can)h(b)q(e)h(preserv)o(ed)e(at)i (program)g(execution)e(time.)35 b(In)22 b(the)f(\014rst)h(step)f(of)h(the)0 255 y(P)o(ath,)g(a)f(scien)o(tist)e(dev)o(elops)h(a)i(mathematical)17 b(mo)q(del)j(for)h(the)g(ph)o(ysical)e(phenomenon.)34 b(Next,)0 315 y(an)17 b(algorithm)e(is)g(dev)o(elop)q(ed)h(to)g(solv)o(e)f(for)i(the)e (relev)m(an)o(t)h(v)m(ariables)g(of)g(the)g(mathematical)d(mo)q(del.)0 376 y(With)i(the)g(algorithm)f(in)h(hand,)h(a)g(program)f(is)g(written.)20 b(Finally)l(,)14 b(the)h(program)g(is)g(compiled)e(and)0 436 y(link)o(ed)i(to)j(pro)q(duce)f(the)g(\014nal)g(instruction)g(stream.)22 b(Ob)o(viously)l(,)16 b(signi\014can)o(t)h(parallelism)e(can)i(b)q(e)0 496 y(lost)k(at)h(eac)o(h)e(step.)36 b(Simplifying)19 b(assumptions)i(in)f (the)h(mathematical)d(mo)q(del)i(ma)o(y)g(eliminate)0 556 y(parallel)14 b(asp)q(ects)i(of)g(the)f(phenomenon,)f(sync)o(hronization)h(p)q(oin)o(ts)h (and)f(sequen)o(tial)f(orderings)i(can)0 616 y(b)q(e)h(in)o(tro)q(duced)f(b)o (y)g(the)g(algorithm,)g(false)g(dep)q(endences)g(and)h(an)o(y)g(n)o(um)o(b)q (er)d(of)j(other)g(pitfalls)f(can)0 677 y(b)q(e)22 b(in)o(tro)q(duced)g(when) g(the)g(program)f(is)h(written,)h(and)f(ev)o(en)f(the)h(b)q(est)g(compilers)e (ma)o(y)g(fail)i(to)0 737 y(detect)15 b(all)h(the)g(parallelism)e(a)o(v)m (ailable)h(in)h(the)g(program.)0 881 y Fe(3.2)70 b(Hardw)n(are)23 b(P)n(arallelism)0 974 y Fm(The)c(hardw)o(are)g(dev)o(elopmen)o(t)d(pro)q (cess)j(starts)h(with)f(a)g(set)f(of)i(system)d(requiremen)o(ts,)f(including) 0 1034 y(those)c(dictated)f(b)o(y)g(non)o(tec)o(hnical)f(criteria)g(suc)o(h)h (as)h(mark)o(et)e(considerations.)20 b(Giv)o(en)10 b(these)h(criteria)0 1094 y(an)16 b(arc)o(hitect)e(dev)o(elops)g(a)i(system)e(arc)o(hitecture,)g (and)i(\014nally)e(a)i(group)h(of)e(engineers)g(dev)o(elops)g(an)0 1154 y(implem)o(en)o(tati)o(on)20 b(of)j(that)f(arc)o(hitecture.)796 1136 y Fj(4)854 1154 y Fm(A)f(giv)o(en)h(arc)o(hitecture)e(ma)o(y)h(b)q(e)h (able)g(to)h(explicitly)0 1214 y(or)c(implicitly)c(supp)q(ort)21 b(parallelism)16 b(in)j(a)h(v)m(ariet)o(y)e(of)h(w)o(a)o(ys,)g(but)h(the)f (ac)o(hiev)m(able)f(parallelism)e(is)0 1275 y(determined)c(b)o(y)i(the)g (particular)g(implem)o(en)o(tation)e(of)j(the)f(arc)o(hitecture.)19 b(In)14 b(this)g(w)o(a)o(y)g(parallelism)0 1335 y(is)i(gained)g(or)h(lost)g (at)f(eac)o(h)g(stage)h(of)f(the)g(hardw)o(are)h(dev)o(elopmen)o(t)c(pro)q (cess.)0 1479 y Fe(3.3)70 b(Realizi)o(ng)20 b(P)n(arallelism)g(at)j(Run)n (time)0 1572 y Fm(Ev)o(en)c(if)g(the)g(soft)o(w)o(are)h(instruction)f(stream) f(and)i(the)f(hardw)o(are)h(arc)o(hitecture)e(implem)o(en)o(tation)0 1632 y(eac)o(h)g(con)o(tain)f(large)h(amoun)o(ts)g(of)g(parallelism,)e(high)i (p)q(erformance)f(will)g(not)h(b)q(e)g(realized)f(unless)0 1692 y(the)f(t)o(w)o(o)h(parallelism)d(con\014gurations)k(are)f(compatible.)j (One)d(common)d(mismatc)o(h)g(is)i(grain)h(size;)0 1752 y(the)k(parallel)f (units)h(of)g(w)o(ork)g(in)g(the)g(soft)o(w)o(are)g(ma)o(y)f(b)q(e)h(to)q(o)h (small)d(to)j(b)q(e)f(executed)f(e\016cien)o(tly)0 1812 y(giv)o(en)d(the)h (parallelism)e(o)o(v)o(erhead)i(presen)o(t)g(in)g(the)g(hardw)o(are.)28 b(In)18 b(other)g(cases)g(the)h(\\shap)q(es")h(of)0 1872 y(the)g(parallelism) e(ma)o(y)g(not)j(matc)o(h.)31 b(F)l(or)20 b(example,)f(a)h(massiv)o(ely)e (parallel)h(arc)o(hitecture)g(with)h(a)0 1933 y(t)o(w)o(o-dimensional)c(grid) i(of)g(pro)q(cessors)h(ma)o(y)d(not)j(b)q(e)f(able)f(to)h(p)q(erform)f (computations)h(along)g(the)0 1993 y(diagonals)f(of)g(matrices)d(at)j(a)g (high)f(rate.)0 2137 y Fe(3.4)70 b(Benc)n(hmarking)22 b(Metho)r(dologies)g (and)i(the)e(P)n(ath)0 2230 y Fm(Eac)o(h)15 b(agen)o(t)h(in)e(the)h(P)o(ath)h (T)l(o)f(P)o(erformance)f(represen)o(ts)g(a)i(scien)o(ti\014c)d(discipline)g (dev)o(oted)i(to)g(min-)0 2290 y(imizing)e(the)i(loss)h(of)f(parallelism.)k (F)l(or)c(example,)e(the)i(Numerical)e(Analysts)i(dev)o(elop)f(alternativ)o (e)0 2350 y(algorithms)23 b(to)h(solv)o(e)f(a)h(giv)o(en)f(mathematical)e(mo) q(del.)43 b(Soft)o(w)o(are)23 b(Engineers)h(study)g(w)o(a)o(ys)g(to)0 2410 y(instan)o(tiate)19 b(a)h(giv)o(en)f(algorithm)f(as)j(programs,)f(etc.) 30 b(Eac)o(h)20 b(of)g(these)f(disciplines)f(mak)o(es)g(trade-)0 2470 y(o\013s)k(with)g(the)f(parallelism)e(against)j(other)f(ob)s(jectiv)o(e) f(criteria)g(suc)o(h)h(as)h(reducing)f(the)g(n)o(um)o(b)q(er)p 0 2514 750 2 v 56 2545 a Fh(4)75 2560 y Fg(The)d(arc)o(hitecture)i(and)e(arc) o(hitecture)h(implemen)o(tation)c(are)j(often)g(confused,)h(but)f(they)h(are) f(distinct.)30 b(F)m(or)0 2610 y(example,)12 b(the)h(IBM)h(370)e(line)h (consists)h(of)f(di\013eren)o(t)h(implemen)o(tatio)o(ns)d(of)i(the)g(same)g (arc)o(hitecture.)19 b(Similarly)m(,)9 b(the)0 2659 y(Cra)o(y)14 b(1,)f(Cra)o(y)g(X-MP)m(,)g(and)h(Cra)o(y)g(Y-MP)g(can)g(b)q(e)g(lo)q(osely)g (view)o(ed)g(as)g(instan)o(tiations)f(of)g(the)h(same)g(arc)o(hitecture.)925 2828 y Fm(5)p eop %%Page: 6 12 bop 0 195 a Fm(of)24 b(op)q(erations.)44 b(The)23 b(\014eld)g(of)h(P)o (erformance)e(Ev)m(aluation)i(attempts)e(to)i(measure)e(parallelism)0 255 y(\(p)q(erformance\))15 b(at)i(particular)f(steps)g(on)h(the)f(P)o(ath.) 73 315 y(When)j(parallelism)c(is)k(measured)e(at)h(t)o(w)o(o)h(di\013eren)o (t)e(steps,)i(w)o(e)f(call)f(the)h(highest)h(and)g(lo)o(w)o(est)0 376 y(measured)13 b(step)h(in)g(the)g(P)o(ath)h(the)f Fi(pr)n(ob)n(e)h(p)n (oints)p Fm(.)20 b(F)l(or)14 b(example,)e(if)i(one)h(uses)f(a)h(program)f(k)o (ernel)f(to)0 436 y(measure)g(the)g(execution)g(time)f(of)i(a)g(section)g(of) g(co)q(de,)g(then)g(the)f(lo)q(cation)h(of)h(the)e(\014rst)h(prob)q(e)h(p)q (oin)o(t)0 496 y(is)f(at)h(the)g(Program)f(lev)o(el,)e(and)k(the)e(second)h (at)g(the)f(Execution)g(Time)e(lev)o(el.)19 b(If)14 b(one)h(measures)e(the)0 556 y(mega\015ops)k(rate)h(of)g(the)f(k)o(ernel,)f(then)h(the)g(second)h (prob)q(e)g(p)q(oin)o(t)g(w)o(ould)f(b)q(e)h(at)g(the)f(P)o(erformance)0 616 y(lev)o(el.)73 677 y(Since)j(the)g(purp)q(ose)i(of)f(an)o(y)f(program)h (is)f(to)h(pro)o(vide)f(a)h(solution)g(to)g(a)g(problem,)e(the)i(most)0 737 y(imp)q(ortan)o(t)c(p)q(erformance)g(measure)g(is)h(time)e(to)i(solution) g(|)g(ho)o(w)h(quic)o(kly)c(the)j(problem)f(can)h(b)q(e)0 797 y(solv)o(ed.)23 b(F)l(or)17 b(this)g(question)f(the)h(appropriate)h(prob)q(e) f(p)q(oin)o(ts)h(are)f(at)g(the)g(Ph)o(ysical)f(Phenomenon)0 857 y(and)21 b(Execution)f(Time)f(lev)o(els.)32 b(Almost)19 b(all)h(b)q(enc)o(hmark)f(activities)g(appro)o(ximate)g(this)i(desired)0 917 y(v)m(alue)c(b)o(y)f(using)h(a)g(di\013eren)o(t)f(\(but)h(represen)o (tativ)o(e\))e(set)i(of)g(b)q(enc)o(hmarks)e(problems)h(and)h(general-)0 978 y(izing)g(the)g(results.)26 b(Some)16 b(sets)i(of)g(b)q(enc)o(hmark)e (programs,)i(called)f Fi(kernel)i Fm(b)q(enc)o(hmarks,)d(con)o(tain)0 1038 y(a)22 b(set)g(of)g(program)g(fragmen)o(ts)e(whose)j(c)o(haracteristics) d(are)i(designed)g(to)g(mimic)c(those)k(of)g(real)0 1098 y(programs.)j(The)18 b(Liv)o(ermore)d(F)l(ortran)j(Kernels)e([1])h(and)h(Linpac)o(k)g(b)q(enc)o (hmark)d([2])i(are)h(examples)0 1158 y(of)d(k)o(ernel)f(b)q(enc)o(hmarks.)19 b(The)d(results)e(of)i(k)o(ernel)d(b)q(enc)o(hmarks)h(are)h(usually)g(rep)q (orted)g(in)g(terms)f(of)0 1218 y(mega\015ops.)24 b(Suc)o(h)17 b(b)q(enc)o(hmarks)f(measure)g(the)h(parallelism)e(exploited)h(at)i (execution)e(time)f(giv)o(en)0 1279 y(a)i(sp)q(eci\014ed)e(amoun)o(t)h(of)g (parallelism)e(\(a)j(\014xed)f(program\))g(at)g(the)g(Program)h(lev)o(el.)73 1339 y(More)j(recen)o(tly)l(,)g(b)q(enc)o(hmarks)f(comp)q(osed)h(of)h(sets)g (of)g(application)f(programs)h(ha)o(v)o(e)e(b)q(ecome)0 1399 y(a)o(v)m(ailable.)25 b(The)17 b(P)o(erfect)g(Benc)o(hmarks)e([3])i(and)h (SPEC)h(Benc)o(hmark)c(Suite)i([4])g(are)h(t)o(w)o(o)f(p)q(opular)0 1459 y(examples.)23 b(Prop)q(onen)o(ts)c(of)f(application)f(b)q(enc)o(hmark)f (sets)i(claim)e(that)i(a)g(set)f(of)h(k)o(ernels)e(cannot)0 1519 y(adequately)i(represen)o(t)g(the)h(p)q(erformance)e(c)o(haracteristics) h(of)h(application)g(programs;)g(only)g(full)0 1579 y(programs)14 b(can)h(adequately)f(test)g(the)g(p)q(erformance)f(c)o(haracteristics)g(of)i (a)f(computer)f(system.)19 b(Re-)0 1640 y(sults)14 b(are)g(usually)f(giv)o (en)g(as)h(execution)f(times)e(for)j(eac)o(h)g(co)q(de)f(in)h(the)f(set,)h (so)g(the)g(prob)q(e)g(p)q(oin)o(ts)g(are)0 1700 y(at)i(the)f(Program)h(and)g (Execution)e(Time)g(lev)o(els.)19 b(Some)14 b(application)i(b)q(enc)o(hmark)e (sets,)h(including)0 1760 y(the)f(P)o(erfect)f(Benc)o(hmarks)g(and)i(the)f (NAS)g(P)o(arallel)f(Benc)o(hmarks)f([5],)i(allo)o(w)g(the)g(user)h(to)g (increase)0 1820 y(p)q(erformance)e(b)o(y)g(mo)q(difying)f(the)i(source)f(co) q(de;)i(c)o(hanges)f(from)e(the)i(simple)d(insertion)i(of)h(compiler)0 1880 y(directiv)o(es)k(to)i(the)g(substitution)g(of)g(new)g(algorithms)f(are) h(allo)o(w)o(ed)f(as)i(long)f(as)h(the)e(correctness)0 1941 y(of)f(the)f(program)g(is)g(preserv)o(ed.)23 b(In)17 b(suc)o(h)g(cases)h(the) f(\014rst)g(prob)q(e)h(p)q(oin)o(t)g(is)f(mo)o(v)o(ed)e(from)h(the)h(Pro-)0 2001 y(gram)i(lev)o(el)e(to)j(the)f(Algorithm)f(or)i(Mathematical)d(Mo)q(del) j(lev)o(el,)d(dep)q(ending)j(on)g(the)f(exten)o(t)g(of)0 2061 y(the)d(mo)q(di\014cations.)73 2121 y(W)l(e)d(th)o(us)h(ha)o(v)o(e)f(sho)o (wn)h(ho)o(w)g(the)f(metho)q(dologies)g(of)h(sev)o(eral)e(p)q(opular)j(b)q (enc)o(hmark)d(sets)h(can)h(b)q(e)0 2181 y(sp)q(eci\014ed)h(in)f(our)i (terminology)l(.)j(No)c(single)g(metho)q(dology)f(is)h(the)g(correct)g(one.) 21 b(Eac)o(h)15 b(b)q(enc)o(hmark)0 2242 y(metho)q(dology)e(measures)f (something)g(di\013eren)o(t)h(and)g(the)g(user)g(needs)g(to)h(decide)e(if)h (the)g(b)q(enc)o(hmark)0 2302 y(tests)h(and)g(measures)f(the)h(desired)f (agen)o(ts.)21 b(Additionally)l(,)12 b(care)i(m)o(ust)e(b)q(e)i(tak)o(en)g (when)g(comparing)0 2362 y(results)j(of)g(di\013eren)o(t)g(b)q(enc)o(hmarks)f (to)h(see)g(that)h(the)f(same)f(set)h(of)g(agen)o(ts)h(is)f(b)q(eing)h (measured.)k(A)0 2422 y(more)15 b(detailed)g(description)h(of)g(these)g(and)h (other)f(b)q(enc)o(hmarks)f(can)i(b)q(e)f(found)h(in)f([6][7)o(][8].)925 2828 y(6)p eop %%Page: 7 13 bop 0 195 a Fe(3.5)70 b(The)28 b(P)n(arallelism)e(Matrix:)43 b(A)29 b(Measure)g(of)h(Executed)e(P)n(aral-)157 270 y(leli)o(sm)0 362 y Fm(An)o(y)16 b(metric)e(that)j(considers)f(only)g(soft)o(w)o(are)h (parallelism)d(\(e.g.)22 b(b)o(y)16 b(assuming)g(in\014nite)g(hardw)o(are)0 422 y(resources\))22 b(is)h(merely)c(a)k(measure)e(of)i(the)f(parallelism)e Fi(p)n(otential)j Fm(con)o(tained)f(in)g(the)g(soft)o(w)o(are.)0 483 y(Corresp)q(ondingly)l(,)k(an)o(y)d(metric)e(based)j(solely)f(on)h(hardw) o(are)g(parallelism)e(describ)q(es)h(only)g(the)0 543 y Fi(p)n(otential)d Fm(parallelism)d(of)j(the)f(hardw)o(are.)807 525 y Fj(5)857 543 y Fm(Of)g(ultimate)f(in)o(terest)g(is)h(the)g(parallelism)e(ac)o(hiev)o (ed)0 603 y(when)g(the)f(soft)o(w)o(are)g(and)i(hardw)o(are)f(parallelism)d (in)o(teract)h(during)i(program)f(execution.)21 b(All)15 b(the)0 663 y(e\013orts)20 b(undertak)o(en)f(to)h(preserv)o(e)e(soft)o(w)o(are)h (parallelism)e(and)j(maximiz)o(e)d(hardw)o(are)i(parallelism)0 723 y(can)14 b(b)q(e)h(view)o(ed)e(as)i(preparation)f(for)h(this)f(ev)o(en)o (t.)19 b(W)l(e)14 b(call)g(the)g(parallelism)d(that)k(results)f(from)f(the)0 784 y(in)o(teraction)i(of)i(hardw)o(are)f(and)h(soft)o(w)o(are)g(parallelism) c(the)j Fi(exe)n(cute)n(d)j(p)n(ar)n(al)r(lelism)p Fm(.)73 844 y(Curren)o(tly)l(,)13 b(commonplace)e(measures)h(of)i(executed)e (parallelism)g(are)h(limited)e(to)j(p)q(ost)h(mortem)0 904 y(a)o(v)o(erages.)34 b(In)20 b(some)f(cases)i(soft)o(w)o(are)f(tec)o(hniques) f(are)i(used)f(to)h(estimate)e(the)h(op)q(eration)h(coun)o(t)0 964 y(in)g(a)g(program)g(or)g(co)q(de)g(segmen)o(t,)f(then)h(this)f(coun)o(t) h(is)g(divided)f(b)o(y)g(the)h(run)o(time)d(to)j(compute)0 1024 y(the)e(a)o(v)o(erage)g(parallelism.)28 b(On)20 b(some)e(systems)g(sp)q (ecial)h(hardw)o(are)h(monitors)e(coun)o(t)i(prede\014ned)0 1084 y(classes)h(of)h(op)q(erations,)h(and)f(this)f(information)f(is)h(used)g (to)h(compute)e(p)q(ost)i(mortem)d(a)o(v)o(erages.)0 1145 y(The)e(hardw)o (are)g(approac)o(h)h(has)f(the)g(p)q(oten)o(tial)f(for)h(greater)g(accuracy)f (but)h(is)g(still)e(limited)f(to)j(the)0 1205 y(simple)d(accum)o(ulation)g (of)j(total)g(op)q(eration)g(coun)o(ts.)73 1265 y(Supp)q(ose)e(that)f(w)o(e)f (ha)o(v)o(e)g(de\014ned)h(\\w)o(ork")g(to)g(b)q(e)g(\015oating)h(p)q(oin)o(t) f(op)q(erations,)h(and)f(consider)g(an)0 1325 y(arbitrary)k(computer)e (system)h(that)h(can)g(collectiv)o(ely)d(compute)h Fk(f)23 b Fm(\015oating)c(p)q(oin)o(t)f(op)q(erations)h(in)0 1385 y(one)c(clo)q(c)o (k)e(p)q(erio)q(d.)22 b(A)14 b(natural)h(extension)f(to)h(the)f(simple)f(p)q (ost)i(mortem)d(a)o(v)o(erage)i(is)h(a)g(histogram,)0 1446 y Fk(W)21 b Fm(=)13 b Ff(h)p Fk(W)183 1453 y Fj(0)204 1446 y Fk(;)8 b(:)g(:)g(:)f(;)h(W)359 1453 y Fc(f)382 1446 y Ff(i)p Fm(,)13 b(where)f Fk(W)611 1453 y Fc(i)638 1446 y Fm(is)g(the)g(n)o(um)o(b)q (er)f(of)i(clo)q(c)o(k)e(p)q(erio)q(ds)j(during)e(whic)o(h)g Fk(i)g Fm(\015oating)i(p)q(oin)o(t)0 1548 y(op)q(erations)20 b(w)o(ere)e(completed)e(sim)o(ultaneously)l(.)27 b(The)19 b(sum)f Fk(t)f Fm(=)1264 1491 y Fc(f)1245 1506 y Fb(X)1246 1597 y Fc(i)p Fj(=0)1313 1548 y Fk(W)1359 1555 y Fc(i)1392 1548 y Fm(is)h(the)h(n)o(um)o(b) q(er)e(of)i(clo)q(c)o(k)0 1690 y(p)q(erio)q(ds)e(consumed)e(b)o(y)h(the)g(en) o(tire)f(program,)g(and)i(the)f(w)o(eigh)o(ted)f(sum)h Fk(w)f Fm(=)1500 1633 y Fc(f)1480 1648 y Fb(X)1482 1739 y Fc(i)p Fj(=0)1549 1690 y Fk(iW)1612 1697 y Fc(i)1642 1690 y Fm(is)h(the)g(total)0 1782 y(amoun)o(t)i(of)i(w)o(ork,)f(in)g(this)g(case)g(\015oating)h(p)q(oin)o (t)g(op)q(erations,)g(p)q(erformed)e(b)o(y)h(the)g(program.)29 b(T)l(o)0 1842 y(facilitate)17 b(comparisons)h(b)q(et)o(w)o(een)f(programs)h (that)h(ha)o(v)o(e)f(di\013eren)o(t)f(execution)g(times,)f(w)o(e)i(divide)0 1902 y(eac)o(h)e(en)o(try)g(in)g(the)g(histogram)g(b)o(y)g Fk(t)p Fm(,)g(the)g(total)h(execution)e(time)f(in)i(clo)q(c)o(k)g(p)q(erio)q (ds,)h(to)g(pro)q(duced)0 1963 y(a)f(normalized)d(histogram)i(called)g(the)g Fi(p)n(ar)n(al)r(lelism)i(ve)n(ctor)f Fk(P)21 b Fm(=)14 b Ff(h)p Fk(P)1276 1970 y Fj(0)1296 1963 y Fk(;)8 b(:)g(:)g(:)g(;)g(P)1437 1970 y Fc(f)1460 1963 y Ff(i)p Fm(,)15 b(where)g Fk(P)1679 1970 y Fc(i)1708 1963 y Fm(=)e Fk(W)1805 1970 y Fc(i)1819 1963 y Fk(=t)p Fm(.)0 2023 y(By)g(construction,)g(eac)o(h)g(en)o(try)f Fk(P)626 2030 y Fc(i)654 2023 y Fm(has)i(a)g(v)m(alue)f(b)q(et)o(w)o(een)f (0.0)h(and)h(1.0)g(that)g(indicates)e(the)h(fraction)0 2083 y(of)k(time)d(during)i(whic)o(h)g Fk(i)g Fm(units)g(of)g(w)o(ork)h(w)o(ere)e (completed)f(in)i(parallel.)73 2143 y(When)11 b(the)g(de\014nition)g(of)g (\\w)o(ork")h(includes)e(op)q(erations)j(of)e(t)o(w)o(o)g(or)g(more)f(t)o(yp) q(es)h(that)h(need)e(to)i(b)q(e)0 2203 y(distinguished,)g(the)f(parallelism)e (v)o(ector)i(is)g(extended)g(to)h(higher)f(dimensions)f(in)h(a)h(straigh)o (tforw)o(ard)0 2264 y(w)o(a)o(y)l(.)21 b(F)l(or)16 b(example)e(supp)q(ose)j (w)o(e)e(de\014ne)h(t)o(w)o(o)g(distinct)f(kinds)g(of)i(w)o(ork,)e(memory)e (op)q(erations)k(and)0 2324 y(\015oating)24 b(p)q(oin)o(t)f(op)q(erations,)i (and)f(consider)f(a)g(mac)o(hine)e(that)i(can)g(p)q(erform)f(up)h(to)h Fk(f)k Fm(\015oating)0 2384 y(p)q(oin)o(t)22 b(and)g Fk(m)g Fm(memory)d(op)q(erations)k(in)e(one)h(clo)q(c)o(k)f(p)q(erio)q(d.)38 b(In)22 b(this)g(case)f(a)i(t)o(w)o(o-dimensional)0 2444 y(histogram)c Fk(W)25 b Fm(=)18 b Fk(W)404 2451 y Fc(ij)435 2444 y Fk(;)8 b Fm(0)18 b Ff(\024)h Fk(i)f Ff(\024)g Fk(f)s(;)8 b Fm(0)18 b Ff(\024)g Fk(j)k Ff(\024)c Fk(m)p Fm(,)h(is)g(used,)g(where)g Fk(W)1342 2451 y Fc(ij)1391 2444 y Fm(is)g(the)f(n)o(um)o(b)q(er)g(of)h(clo)q (c)o(k)0 2504 y(p)q(erio)q(ds)13 b(during)f(whic)o(h)f Fk(i)g Fm(\015oating)i(p)q(oin)o(t)f(op)q(erations)h(and)f Fk(j)j Fm(memory)9 b(op)q(erations)k(w)o(ere)e(completed)p 0 2548 750 2 v 56 2579 a Fh(5)75 2594 y Fg(The)j(most)f(notorious)g(example,)f(of)i (course,)g(is)g(p)q(eak)g(mega\015ops.)925 2828 y Fm(7)p eop %%Page: 8 14 bop 0 238 a Fm(sim)o(ultaneously)l(.)19 b(As)d(b)q(efore)h(w)o(e)f(divide)f (b)o(y)g(the)h(total)h(n)o(um)o(b)q(er)d(of)j(clo)q(c)o(k)e(p)q(erio)q(ds)i Fk(t)d Fm(=)1681 181 y Fc(f)1661 196 y Fb(X)1662 287 y Fc(i)p Fj(=0)1744 184 y Fc(m)1730 196 y Fb(X)1729 287 y Fc(j)r Fj(=0)1799 238 y Fk(W)1845 245 y Fc(ij)0 339 y Fm(to)e(obtain)g(a)g(normalized)e (histogram)h(called)g(the)g Fi(p)n(ar)n(al)r(lelism)j(matrix)d Fk(P)c Fm(,)12 b(where)g(eac)o(h)f Fk(P)1647 346 y Fc(ij)1691 339 y Fm(=)j Fk(W)1789 346 y Fc(ij)1819 339 y Fk(=t)p Fm(.)73 399 y(In)h(a)h(similar)e(w)o(a)o(y)h(parallelism)e(matrices)h(of)i(arbitrary) f(dimension)f(can)i(b)q(e)g(constructed)f(with)0 460 y(one)20 b(dimension)e(for)i(eac)o(h)f(of)h(the)f(di\013eren)o(t)g(kinds)g(of)h(w)o (ork)g(that)g(are)f(of)h(in)o(terest.)30 b(Some)19 b(other)0 520 y(p)q(ossibilities)11 b(for)i(\\w)o(ork")g(include)e(logical)h(op)q (erations,)h(in)o(teger)f(op)q(erations,)h(and)g(I/O)f(op)q(erations.)0 580 y(Dep)q(ending)24 b(on)g(ho)o(w)g(w)o(e)f(de\014ne)g(w)o(ork)g(w)o(e)h (can)f(obtain)h(v)m(arious)g(parallelism)e(pro\014les)h(from)g(a)0 640 y(program's)18 b(parallelism)f(matrix.)26 b(In)19 b(the)f(t)o(w)o(o)h (dimensional)e(case)i(describ)q(ed)f(ab)q(o)o(v)o(e,)g(the)h(v)o(ector)0 700 y(of)h(ro)o(w)g(sums)f(of)h(the)g(parallelism)d(matrix)h(giv)o(es)h(a)h (pro\014le)g(of)g(the)f(memory)e(parallelism,)h(while)0 761 y(the)f(v)o(ector)f(of)h(column)f(sums)g(sho)o(ws)i(the)f(pro\014le)g(of)g (\015oating-p)q(oin)o(t)h(parallelism.)k(The)17 b(v)o(ector)f(of)0 821 y(diagonal)i(sums)d(represen)o(ts)h(a)i(pro\014le)e(of)h(w)o(ork)g (parallelism)d(where)i(\015oating)i(p)q(oin)o(t)f(and)g(memory)0 881 y(op)q(erations)j(are)f(not)g(distinguished)g(\(i.e.)28 b(the)18 b(unit)h(of)g(w)o(ork)g(is)g(de\014ned)g(as)g(a)g(\015oating-p)q (oin)o(t)i Fi(or)0 941 y Fm(memory)13 b(op)q(eration\).)0 1086 y Fe(3.6)70 b(Quan)n(tifying)22 b(Di\013erences)e(in)i(Executed)h(P)n (arallelism)0 1178 y Fm(Although)c(similar)d(programs)i(will)f(exhibit)h (similar)e(a)o(v)o(erage)i(parallelism)e(on)i(a)h(giv)o(en)f(mac)o(hine,)0 1238 y(the)e(con)o(v)o(erse)f(is)g(not)i(true;)e(similar)f(a)o(v)o(erage)i (parallelism)d(do)q(es)k(not)f(indicate)f(that)i(programs)f(are)0 1298 y(similar)c([12].)20 b(P)o(arallelism)12 b(v)o(ectors)h(and)i(matrices)d (rev)o(eal)h(di\013erences)g(in)h(program)g(b)q(eha)o(vior)g(that)0 1359 y(cannot)22 b(b)q(e)g(distinguished)f(b)o(y)g(a)o(v)o(erages.)36 b(W)l(e)22 b(can)f(compare)g(the)g(parallelism)e(pro\014les)i(of)h(t)o(w)o(o) 0 1419 y(programs)d(b)o(y)f(comparing)g(the)h(parallelism)e(matrices)g(for)i (eac)o(h)f(program)h(using)g(the)g(F)l(rob)q(enius)0 1479 y(matrix)e(norm)h (to)i(quan)o(tify)e(the)g(di\013erence.)28 b(If)19 b Fk(A)f Fm(is)h(the)g(t)o(w)o(o-dimensional)e Fk(m)c Ff(\002)f Fk(n)19 b Fm(parallelism)0 1539 y(matrix)f(for)j(Program)f(1)h(and)f Fk(B)j Fm(is)d(the)g Fk(m)14 b Ff(\002)f Fk(n)20 b Fm(parallelism)e(matrix)h (for)h(Program)g(2,)h(then)f(w)o(e)0 1599 y(gauge)d(the)f(di\013erence)f(in)h (parallelism)e(b)q(et)o(w)o(een)h(the)h(t)o(w)o(o)h(programs)f(b)o(y)426 1709 y(di\013erence)f(in)h(parallelism)39 b(=)i Ff(k)p Fk(A)11 b Ff(\000)g Fk(B)s Ff(k)1239 1724 y Fc(F)973 1885 y Fm(=)1052 1802 y Fb(v)1052 1825 y(u)1052 1850 y(u)1052 1875 y(t)p 1096 1802 353 2 v 1111 1831 a Fc(m)1096 1844 y Fb(X)1097 1935 y Fc(i)p Fj(=1)1184 1831 y Fc(n)1165 1844 y Fb(X)1164 1935 y Fc(j)r Fj(=1)1234 1885 y Ff(j)p Fk(a)1274 1892 y Fc(ij)1315 1885 y Ff(\000)g Fk(b)1386 1892 y Fc(ij)1416 1885 y Ff(j)1430 1871 y Fj(2)1813 1885 y Fm(\(1\))0 2038 y(In)o(tuitiv)o(ely)l(,)j(the)j(F)l (rob)q(enius)g(norm)f(represen)o(ts)h(the)g(\\distance")h(b)q(et)o(w)o(een)e (t)o(w)o(o)h(matrices,)e(just)i(as)0 2098 y(the)e(Euclidean)f(form)o(ula)g (is)g(used)h(to)h(measure)d(the)i(distance)g(b)q(et)o(w)o(een)f(t)o(w)o(o)h (p)q(oin)o(ts.)21 b(This)15 b(metho)q(d)0 2159 y(is)h(used)g(to)h(compare)e (some)g(application)h(b)q(enc)o(hmarks)f(in)h(Section)g(5.)73 2219 y(Recall)j(that)h(the)g(matrix)f(elemen)o(ts)e(represen)o(t)i(the)h (fraction)g(of)g(the)g(normalized)e(execution)0 2279 y(time,)g(so)j(that)f (their)f(sum)g(equals)g(1.0.)32 b(This)20 b(means)f(that)h(the)g (di\013erence)f(in)g(parallelism)f(ma)o(y)0 2339 y(range)i(from)f(0.0)h(for)h (t)o(w)o(o)e(programs)h(with)g(iden)o(tical)e(parallelism)g(distributions,)i (to)1661 2298 y Ff(p)p 1702 2298 25 2 v 1702 2339 a Fm(2)h(in)e(the)0 2399 y(case)d(where)g(eac)o(h)g(matrix)e(has)j(only)f(one)h(non-zero)f (elemen)o(t)d(\(with)j(v)m(alue)g(1.0\))h(and)g(the)f(lo)q(cation)0 2460 y(of)h(the)f(non-zero)g(elemen)o(t)e(in)h(eac)o(h)h(matrix)f(is)h (di\013eren)o(t.)925 2828 y(8)p eop %%Page: 9 15 bop 0 203 a Fl(4)83 b(Rederiv)-5 b(ation)24 b(of)i(Basic)e(P)n(arallel)i(P)n (erformance)e(Met-)124 295 y(rics)0 404 y Fm(This)d(section)g(striv)o(es)f (to)h(justify)g(the)f(approac)o(h)i(to)g(Benc)o(hmarking)c(describ)q(ed)j(in) f Ff(x)p Fm(5)i(and)g(our)0 464 y(emphasis)16 b(on)h(parallelism)d(b)o(y)j (sho)o(wing)g(that)g(man)o(y)e(common)g(p)q(erformance)h(metrics,)e (including)0 525 y(Amdahl's)g(La)o(w,)j(are)f(directly)f(related)g(to)i(the)f (parallelism)e(in)i(the)g(hardw)o(are)g(and)h(soft)o(w)o(are.)0 669 y Fe(4.1)70 b(P)n(erformance)0 761 y Fm(The)16 b(ultimate)e(metric)f(for) k(measuring)e(the)g(p)q(erformance)g(of)i(a)f(mac)o(hine)e(executing)h(a)h (giv)o(en)f(pro-)0 822 y(gram)h(with)g(a)h(\014xed)f(dataset)h(is)f(the)g(in) o(v)o(erse)f(of)i(the)f(w)o(all)g(clo)q(c)o(k)f(time,)f Fi(WCT)p 1473 822 15 2 v 17 w(se)n(c)n(onds)p Fm(,)i(measured)0 882 y(in)g(units)g(of)h(recipro)q(cal)e(seconds,)669 862 y Fj(1)p 653 870 48 2 v 653 899 a Fc(sec)706 882 y Fm(.)618 1019 y Fi(Performanc)n(e)e Fm(=)1093 985 y Fi(1)p 959 1007 294 2 v 959 1053 a(WCT)p 1081 1053 15 2 v 17 w(se)n(c)n(onds)1813 1019 y Fm(\(2\))73 1148 y(If)18 b(a)g(program)g(tak)o(es)g(10)g(elapsed)g(seconds)g(to)h(run,)f(then) g(its)g(p)q(erformance)e(is)1602 1128 y Fj(1)p 1593 1136 36 2 v 1593 1165 a(10)1654 1128 y(1)p 1638 1136 48 2 v 1638 1165 a Fc(sec)1709 1148 y Fm(for)j(that)0 1208 y(problem.)26 b(Another)18 b(program)g(that)h(solv)o(es)e(the)h(same)g(problem)e(has)j(higher)f(p)q (erformance)g(if)f(its)0 1268 y(execution)23 b(time)e(is)i(smaller.)42 b(W)l(e)23 b(clarify)g(the)g(factors)h(that)g(con)o(tribute)f(to)h(p)q (erformance)f(b)o(y)0 1329 y(m)o(ultiplyi)o(ng)14 b(equation)i(\(2\))h(t)o (wice)e(b)o(y)h(a)g(factor)h(whose)g(v)m(alue)f(is)g(1.)369 1511 y Fi(Performanc)n(e)41 b Fm(=)900 1477 y(1)p 766 1499 294 2 v 766 1545 a Fi(WCT)p 888 1545 15 2 v 17 w(se)n(c)n(onds)1075 1511 y Ff(\002)1130 1477 y Fi(work)p 1130 1499 101 2 v 1130 1545 a(work)1246 1511 y Ff(\002)1301 1477 y Fi(total)p 1398 1477 15 2 v 18 w(CPs)p 1301 1499 201 2 v 1301 1545 a(total)p 1398 1545 15 2 v 18 w(CPs)681 1669 y Fm(=)804 1635 y(1)p 766 1657 101 2 v 766 1703 a Fi(work)882 1669 y Ff(\002)987 1635 y Fi(work)p 937 1657 201 2 v 937 1703 a(total)p 1034 1703 15 2 v 18 w(CPs)1153 1669 y Ff(\002)1254 1635 y Fi(total)p 1351 1635 V 19 w(CPs)p 1208 1657 294 2 v 1208 1703 a(WCT)p 1330 1703 15 2 v 17 w(se)n(c)n(onds)1813 1669 y Fm(\(3\))681 1824 y(=)804 1790 y(1)p 766 1812 101 2 v 766 1858 a Fi(work)882 1824 y Ff(\002)11 b Fk(par)q(al)q(l)q(el)q(ism)e Ff(\002)h Fk(cl)q(ock)p 1354 1824 15 2 v 20 w(r)q(ate)354 b Fm(\(4\))16 1946 y(where)73 2048 y Ff(\017)24 b Fi(work)16 b Fm(is)g(the)g(total)h(n)o (um)o(b)q(er)d(of)j(\\op)q(erations")h(in)e(the)g(program.)73 2150 y Ff(\017)24 b Fi(op)n(er)n(ation)15 b Fm(The)h(de\014nition)g(of)g(op)q (eration)h(is)f Fi(arbitr)n(ary)p Fm(.)j(An)d(op)q(eration)h(ma)o(y)e(b)q(e)h (a)g(mem)o(b)q(er)122 2210 y(of)k(the)f(set)h(of)g(all)f(op)q(eration)h(t)o (yp)q(es,)g(only)f(\015oating)i(p)q(oin)o(t)e(op)q(erations,)i(only)f(in)o (teger)e(and)122 2270 y(logical)11 b(op)q(erations,)h(only)f(memory)d(op)q (erations,)13 b(only)e(I/O)g(op)q(erations)h(or)g(an)o(y)f(com)o(bination)122 2330 y(of)16 b(op)q(eration)g(classes)g(that)g(the)f(user)g(ma)o(y)f(wish)i (to)g(de\014ne)f(to)h(suit)f(his)h(purp)q(ose.)22 b(The)15 b(only)122 2391 y(limitation)j(is)i(that)h(the)f(same)g(de\014nition)g(m)o (ust)f(b)q(e)i(used)f(in)g(b)q(oth)i(the)e(\014rst)h(and)g(second)122 2451 y(factors)c(in)f(equation)g(\(3\).)73 2552 y Ff(\017)24 b Fi(total)p 219 2552 V 19 w(CPs)g Fm(is)g(the)g(total)g(n)o(um)o(b)q(er)e (of)j(elapsed)f(clo)q(c)o(k)f(p)q(erio)q(ds)i(required)e(to)i(execute)d(the) 122 2613 y(program.)45 b(It)24 b(has)h(the)f(same)f(v)m(alue)h(as)h(the)f(w)o (all)g(clo)q(c)o(k)f(time,)h(W)o(CT)p 1530 2613 V 18 w(seconds,)i(but)e(is) 122 2673 y(expressed)16 b(in)g(units)g(of)g(clo)q(c)o(k)g(p)q(erio)q(ds.)925 2828 y(9)p eop %%Page: 10 16 bop 73 195 a Ff(\017)24 b Fi(p)n(ar)n(al)r(lelism)19 b Fm(is)f(the)h(ratio)g (of)f(the)h(w)o(ork)f(p)q(erformed)g(to)h(the)f(elapsed)h(time,)d(and)j(is)g (the)f Fi(av-)122 255 y(er)n(age)j Fm(n)o(um)o(b)q(er)f(of)i(op)q(erations)g (pro)q(duced)g(p)q(er)f(clo)q(c)o(k)g(p)q(erio)q(d.)37 b(It)20 b(can)i(include)e(b)q(oth)i(the)122 315 y(concurren)o(t)17 b(execution)g(of)h(m)o(ultiple)d(pro)q(cessors,)k(and)g(the)f(o)o(v)o(erlapp) q(ed)f(execution)g(of)h(sev-)122 376 y(eral)e(instructions)g(within)g(a)g (single)g(pro)q(cessor.)73 477 y Ff(\017)24 b Fi(clo)n(ck)p 226 477 15 2 v 19 w(r)n(ate)15 b Fm(is)f(the)h(ratio)g(of)h(the)f(n)o(um)o(b) q(er)e(of)i(elapsed)g(clo)q(c)o(k)f(p)q(erio)q(ds)i(to)f(the)g(elapsed)f (time)f(in)122 538 y(seconds.)22 b(It)15 b(is)h(the)g(n)o(um)o(b)q(er)f(of)h (clo)q(c)o(k)g(p)q(erio)q(ds)h(in)f(one)g(second.)0 639 y(This)e (de\014nition)f(of)i(p)q(erformance)d(is)i(similar)e(to)i(that)g(giv)o(en)f (in)h([9])f(except)g(that)h(w)o(e)g(use)g(w)o(all)f(clo)q(c)o(k)0 699 y(time)h(rather)i(than)h(CPU)f(time.)0 844 y Fe(4.2)70 b(Execution)22 b(Rate)0 936 y Fm(Although)d(P)o(erformance)d(is)j(the)f (metric)e(of)j(c)o(hoice,)e(another)i(metric,)d Fi(Exe)n(cution)p 1571 936 V 20 w(R)n(ate)p Fm(,)i(is)g(often)0 996 y(used)24 b(b)q(ecause)g(it)f(is)g(easy)h(to)g(compare)f(this)g(metric)f(with)h(the)h Fi(Pe)n(ak)p 1379 996 V 18 w(Exe)n(cution)p 1603 996 V 19 w(R)n(ate)g Fm(metric)0 1057 y(describ)q(ed)16 b(b)q(elo)o(w.)21 b(Execution)p 590 1057 V 17 w(Rate)16 b(ma)o(y)f(b)q(e)h(expressed)g(in)g(sev)o(eral)f (forms.)275 1167 y Fi(Exe)n(cution)p 484 1167 V 19 w(R)n(ate)41 b Fm(=)g Fi(Performanc)n(e)11 b Ff(\002)g Fi(work)638 1261 y Fm(=)819 1228 y Fi(work)p 722 1250 294 2 v 722 1295 a(WCT)p 844 1295 15 2 v 18 w(se)n(c)n(onds)638 1378 y Fm(=)717 1317 y Fb(\022)791 1344 y Fm(1)p 753 1366 101 2 v 753 1412 a Fi(work)869 1378 y Ff(\002)g Fi(p)n(ar)n(al)r(lelism)g Ff(\002)g Fi(clo)n(ck)p 1311 1378 15 2 v 19 w(r)n(ate)1409 1317 y Fb(\023)1450 1378 y Ff(\002)g Fi(work)638 1469 y Fm(=)41 b Fi(p)n(ar)n(al)r(lelism)12 b Ff(\002)f Fi(clo)n(ck)p 1110 1469 V 19 w(r)n(ate)605 b Fm(\(5\))0 1593 y(Execution)p 217 1593 V 17 w(Rate)22 b(is)g(measured)f(in)h(units)g(of) 928 1569 y(op)q(erations)p 890 1582 298 2 v 890 1618 a(W)o(CT)p 1012 1618 15 2 v 18 w(seconds)1192 1593 y(.)39 b(While)22 b(the)g(P)o (erformance)e(mea-)0 1653 y(suremen)o(ts)e(for)i(eac)o(h)g(of)g(t)o(w)o(o)g (programs)g(solving)g(the)g(same)e(problem)h(are)h(alw)o(a)o(ys)g (comparable,)0 1714 y(the)h(Execution)p 306 1714 V 17 w(Rates)g(for)g(t)o(w)o (o)f(programs)h(solving)g(the)g(same)e(problem)h(are)h(only)f(comparable)0 1774 y(if)g(w)o(ork)h(is)f(the)h(same)f(for)h(b)q(oth)g(programs.)35 b(The)21 b(same)f(relationships)g(hold)h(when)g(comparing)0 1834 y(mac)o(hines)16 b(solving)j(the)f(same)f(problem.)26 b(Execution)p 1021 1834 V 17 w(Rate)18 b(is)h(often)f(mistak)o(enly)e (confused)i(with)0 1894 y(p)q(erformance.)0 2039 y Fe(4.3)70 b(P)n(eak)23 b(Execution)f(Rate)0 2131 y Fm(W)l(e)16 b(de\014ne)g(P)o(eak)p 332 2131 V 17 w(Execution)p 563 2131 V 17 w(Rate)g(as)113 2251 y Fi(Pe)n(ak)p 216 2251 V 18 w(Exe)n(cution)p 440 2251 V 19 w(R)n(ate)41 b Fm(=)h Fi(work)11 b Ff(\002)835 2190 y Fb(\022)908 2217 y Fm(1)p 870 2239 101 2 v 870 2285 a Fi(work)987 2251 y Ff(\002)f Fi(maximal)p 1218 2251 15 2 v 19 w(p)n(ar)n(al)r(lelism)h Ff(\002)g Fi(clo)n(ck)p 1626 2251 V 19 w(r)n(ate)1724 2190 y Fb(\023)594 2339 y Fm(=)42 b Fi(maximal)p 856 2339 V 18 w(p)n(ar)n(al)r (lelism)11 b Ff(\002)g Fi(clo)n(ck)p 1263 2339 V 19 w(r)n(ate)0 2449 y Fm(where)j Fi(maximal)p 321 2449 V 18 w(p)n(ar)n(al)r(lelism)g Fm(is)g(the)f(maxim)o(um)d(n)o(um)o(b)q(er)i(of)i(\\op)q(eration")i(results)d (that)i(the)e(hard-)0 2509 y(w)o(are)f(can)h(pro)q(duce)f(in)g(one)h(clo)q(c) o(k)e(p)q(erio)q(d,)i(assuming)f(that)h(all)f(other)g(resources)g(suc)o(h)g (as)h(functional)0 2570 y(units,)18 b(registers)g(and)h(data)g(paths,)g(and)f (of)h(course,)f(the)g(op)q(erands,)h(are)g(alw)o(a)o(ys)e(a)o(v)m(ailable)h (when)0 2630 y(needed.)913 2828 y(10)p eop %%Page: 11 17 bop 73 195 a Fm(As)17 b(an)g(example,)d(a)j(CRA)l(Y)f(C-90)h(with)g(eac)o(h)f (CPU)h(con)o(taining)f(an)h(add)g(and)h(m)o(ultiply)13 b(func-)0 255 y(tional)k(unit,)e(3)i(memory)d(p)q(orts,)j(and)g(2)g(logical)f (functional)h(units,)f(eac)o(h)g(capable)g(of)h(pro)q(ducing)h(2)0 315 y(results)e(p)q(er)g(clo)q(c)o(k)g(p)q(erio)q(d,)g(16)h(CPUs,)f(and)h(a)f (4ns)h(clo)q(c)o(k)f(p)q(erio)q(d)g(has)h(P)o(eak)p 1439 315 15 2 v 18 w(Execution)p 1671 315 V 17 w(Rates)f(of)254 417 y(4)h(\(+,*\))f(results)p 254 433 310 2 v 375 469 a(CP)233 563 y(6)g(memory)e(w)o(ords)p 233 576 353 2 v 375 612 a(CP)243 707 y(4)j(logical)f(results)p 243 720 332 2 v 375 756 a(CP)611 381 y Fb(9)611 418 y(>)611 431 y(>)611 443 y(>)611 456 y(>)611 468 y(>)611 481 y(>)611 493 y(>)611 505 y(>)611 518 y(>)611 530 y(=)611 605 y(>)611 618 y(>)611 630 y(>)611 642 y(>)611 655 y(>)611 667 y(>)611 680 y(>)611 692 y(>)611 705 y(>)611 717 y(;)659 580 y Ff(\002)11 b Fm(16)17 b(CPUs)11 b Ff(\002)g Fm(250)17 b(MHz)c(=)1217 393 y Fb(8)1217 431 y(>)1217 443 y(>)1217 456 y(>)1217 468 y(>)1217 481 y(>)1217 493 y(>)1217 505 y(>)1217 518 y(>)1217 530 y(<)1217 605 y(>)1217 618 y(>)1217 630 y(>)1217 642 y(>)1217 655 y(>)1217 667 y(>)1217 680 y(>)1217 692 y(>)1217 705 y(:)1291 419 y Fm(16.0)k(G\015ops)p 1279 432 278 2 v 1279 469 a(W)o(CT)p 1401 469 15 2 v 18 w(second)1604 444 y(or)1305 563 y(24.0)g(GWs)p 1279 571 278 2 v 1279 607 a(W)o(CT)p 1401 607 15 2 v 18 w(second)1604 583 y(or)1289 701 y(16.0)g(GLops)p 1279 714 278 2 v 1279 751 a(W)o(CT)p 1401 751 15 2 v 18 w(second)0 844 y(where)f Fi(GL)n(ops/WCT)p 425 844 V 16 w(se)n(c)n(ond)g Fm(is)g(one)g(billion)f(logical)h(op)q(erations)i(p)q(er)e(W)o(CT)p 1475 844 V 18 w(second.)21 b(The)0 904 y(P)o(eak)p 107 904 V 17 w(Execution)p 338 904 V 17 w(Rate)15 b(of)g(a)h(mac)o(hine)d(ma)o(y)g(b) q(e)i(ac)o(hiev)m(able)f(for)h(short)h(p)q(erio)q(ds)f(of)h(time,)c(but)k(ma) o(y)0 964 y(not)k(b)q(e)f(sustainable)h(o)o(v)o(er)e(longer)i(p)q(erio)q(ds)g (since)e(the)h(assumptions)h(made)e(in)h(the)g(de\014nition)g(of)0 1024 y(maximal)p 189 1024 V 14 w(parallelism)14 b(are)j(usually)f(not)g(alw)o (a)o(ys)g(true.)0 1168 y Fe(4.4)70 b(Sp)r(eedup)0 1260 y Fm(W)l(e)22 b(de\014ne)f(Sp)q(eedup)h(of)g(a)g(new)g(program)g(o)o(v)o(er)f(an)h(old)g (program)f(that)i(b)q(oth)f(solv)o(e)f(the)h(same)0 1320 y(problem)15 b(as)i(the)f(ratio)g(of)h(the)f(p)q(erformance)f(of)h(eac)o(h)g(program.)620 1448 y Fi(Sp)n(e)n(e)n(dup)e Fm(=)857 1414 y Fi(Performanc)n(e\(new\))p 857 1436 393 2 v 867 1482 a(Performanc)n(e\(old\))0 1578 y Fm(F)l(rom)h(\(2\))i(it)e(follo)o(ws)i(that)609 1666 y Fi(Sp)n(e)n(e)n(dup)d Fm(=)855 1632 y Fi(WCT)p 977 1632 15 2 v 18 w(se)n(c)n(onds\(old\))p 845 1654 416 2 v 845 1700 a(WCT)p 967 1700 15 2 v 18 w(se)n(c)n(onds\(new\))0 1778 y Fm(whic)o(h)e(is)g(related)g(to)g(the)h(standard)g(de\014nition,)g Fk(S)s Fm(\()p Fk(p)p Fm(\))h(=)g Fk(T)7 b Fm(\(1\))p Fk(=T)g Fm(\()p Fk(p)p Fm(\),)12 b(for)h(m)o(ultipro)q(cessor)e(sp)q(eedup)0 1838 y(using)18 b Fk(p)g Fm(pro)q(cessors)h([10].)25 b(Ho)o(w)o(ev)o(er,)16 b(w)o(e)h(can)h(use)f(the)h(de\014nition)f(of)h(P)o(erformance)e(giv)o(en)g (in)i(\(3\))0 1898 y(to)f(exp)q(ose)f(the)g(con)o(tributing)g(factors)h(to)f (Sp)q(eedup)h(in)f(greater)g(detail.)303 2065 y Fi(Sp)n(e)n(e)n(dup)e Fm(=)643 1987 y Fi(1)p 545 1995 223 2 v 545 2034 a(work\(new\))783 2007 y Ff(\002)d Fi(p)n(ar)n(al)r(lelism\(new\))h Ff(\002)f Fi(clo)n(ck)p 1347 2007 15 2 v 19 w(r)n(ate\(new\))p 540 2053 1028 2 v 663 2091 a(1)p 574 2099 203 2 v 574 2138 a(work\(old\))793 2110 y Ff(\002)g Fi(p)n(ar)n(al)r(lelism\(old\))g Ff(\002)g Fi(clo)n(ck)p 1337 2110 15 2 v 19 w(r)n(ate\(old\))0 2218 y Fm(or)298 2293 y Fi(Sp)n(e)n(e)n(dup)j Fm(=)545 2260 y Fi(work\(old\))p 535 2282 223 2 v 535 2328 a(work\(new\))773 2293 y Ff(\002)828 2260 y Fi(p)n(ar)n(al)r(lelism\(new\))p 828 2282 350 2 v 838 2328 a(p)n(ar)n(al)r(lelism\(old\))1193 2293 y Ff(\002)1248 2260 y Fi(clo)n(ck)p 1352 2260 15 2 v 19 w(r)n(ate\(new\))p 1248 2282 324 2 v 1258 2328 a(clo)n(ck)p 1362 2328 15 2 v 19 w(r)n(ate\(old\))1813 2293 y Fm(\(6\))0 2405 y(Equation)21 b(\(6\))g(illustrates)f(the)g(t)o(w)o(o)g(additional)h(factors,)g(the)g(w)o (ork)f(ratio)h(and)g(the)f(parallelism)0 2466 y(ratio,)c(that)h(prev)o(en)o (t)e(accurate)h(mac)o(hine)e(comparisons)i(based)g(solely)g(on)g(the)g(clo)q (c)o(k)p 1606 2466 V 17 w(rate)g(ratio.)73 2526 y(If)21 b(w)o(e)g(are)h (comparing)e(t)o(w)o(o)i(executions)e(on)i(the)g(same)e(mac)o(hine,)g(then)i Fi(clo)n(ck)p 1596 2526 V 19 w(r)n(ate\(new\))h(=)0 2586 y(clo)n(ck)p 104 2586 V 19 w(r)n(ate\(old\))p Fm(.)496 2674 y Fi(Sp)n(e)n(e)n(dup)13 b Fm(=)742 2640 y Fi(work\(old\))p 732 2662 223 2 v 732 2708 a(work\(new\))970 2674 y Ff(\002)1025 2640 y Fi(p)n(ar)n(al)r(lelism\(new\))p 1025 2662 350 2 v 1035 2708 a(p)n(ar)n(al)r(lelism\(old\))1813 2674 y Fm(\(7\))913 2828 y(11)p eop %%Page: 12 18 bop 0 195 a Fm(where)16 b(the)g(\014rst)g(factor)h(is)f(the)g(recipro)q(cal)g (of)g(the)g(Redundancy)l(,)654 316 y Fi(R)n(e)n(dundancy)e Fm(=)980 282 y Fi(work\(new\))p 980 305 223 2 v 990 350 a(work\(old\))1207 316 y Fk(:)0 476 y Fe(4.5)70 b(E\016ciency)0 568 y Fm(W)l(e)19 b(de\014ne)g(the)g(E\016ciency)f(of)i(a)g(new)f(program)g(as)h(the)f(ratio)h (of)g(the)f(realized)f(Sp)q(eedup)h(to)h(the)0 629 y(maxim)o(um)12 b(p)q(ossible)k(Sp)q(eedup.)614 747 y Fi(E\016ciency)f Fm(=)992 713 y Fi(Sp)n(e)n(e)n(dup)p 893 735 363 2 v 893 781 a(maximal)p 1075 781 15 2 v 19 w(Sp)n(e)n(e)n(dup)1813 747 y Fm(\(8\))0 869 y(where,)g(using)i(\(7\),)314 992 y Fi(maximal)p 496 992 V 18 w(Sp)n(e)n(e)n(dup)d Fm(=)803 959 y Fi(work\(old\))p 748 981 314 2 v 748 1026 a(work\(max)p 958 1026 15 2 v 18 w(p)n(ar\))1077 992 y Ff(\002)1131 959 y Fi(maximal)p 1313 959 V 19 w(p)n(ar)n(al)r(lelism)p 1131 981 425 2 v 1179 1026 a(p)n(ar)n(al)r(lelism\(old\))0 1117 y Fm(and)i Fi(work\(max)p 304 1117 15 2 v 18 w(p)n(ar\))e Fm(is)g(the)h(n)o(um)o(b)q(er)e(of)i(op)q(erations)h(in)f(the)g Fi(maximal)p 1339 1117 V 18 w(p)n(ar)n(al)r(lel)24 b Fm(implem)o(e)o(n)o (tation.)73 1177 y(De\014nition)14 b(\(8\))h(is)f(related)g(to)h(the)f (standard)h(de\014nition,)f Fk(E)s Fm(\()p Fk(p)p Fm(\))h(=)e Fk(S)s Fm(\()p Fk(p)p Fm(\))p Fk(=p)p Fm(,)j(for)f(m)o(ultipro)q(cessor)0 1237 y(E\016ciency)g(using)h Fk(p)h Fm(pro)q(cessors)g([10)q(].)k(Ho)o(w)o (ev)o(er,)14 b(w)o(e)i(can)g(use)g(the)g(de\014nition)g(of)h(Sp)q(eedup)f(in) g(\(7\),)0 1297 y(the)i(constan)o(t)h(clo)q(c)o(k)p 396 1297 V 17 w(rate)f(case,)h(to)g(exp)q(ose)g(the)f(con)o(tributing)g(factors)h(to)g (E\016ciency)e(in)h(greater)0 1357 y(detail.)358 1581 y Fi(E\016ciency)43 b Fm(=)791 1495 y Fi(work\(old\))p 781 1511 223 2 v 781 1550 a(work\(new\))1020 1523 y Ff(\002)1075 1495 y Fi(p)n(ar)n(al)r(lelism\(new\)) p 1075 1511 350 2 v 1084 1550 a(p)n(ar)n(al)r(lelism\(old\))p 693 1569 819 2 v 754 1611 a(work\(old\))p 698 1627 314 2 v 698 1666 a(work\(max)p 908 1666 15 2 v 19 w(p)n(ar\))1028 1639 y Ff(\002)1082 1614 y Fi(maximal)p 1264 1614 V 19 w(p)n(ar)n(al)r(lelism)p 1082 1627 425 2 v 1130 1666 a(p)n(ar)n(al)r(lelism\(old\))609 1783 y Fm(=)731 1749 y Fi(p)n(ar)n(al)r(lelism\(new\))p 693 1771 V 693 1817 a(maximal)p 875 1817 15 2 v 19 w(p)n(ar)n(al)r(lelism)1134 1783 y Ff(\002)1189 1749 y Fi(work\(max)p 1399 1749 V 18 w(p)n(ar\))p 1189 1771 314 2 v 1234 1817 a(work\(new\))1813 1783 y Fm(\(9\))0 1905 y(where)17 b(w)o(e)g(observ)o(e)f(the)h(recipro)q(cal)g(of)g(the)g (Redundancy)g(as)h(the)f(second)g(factor,)g(and)h(note)g(that)0 1965 y(E\016ciency)d(is)h(not)g(a)h(function)f(of)g(the)g(p)q(erformance)g (of)g(the)g(old)g(program.)0 2107 y Fe(4.6)70 b(Utili)o(zati)o(on)0 2199 y Fm(Our)13 b(de\014nition)f(of)h(the)g(Utilization)e(of)i(a)h(new)e (program)h(follo)o(ws)g(from)f(\(9\))h(and)g(from)f(the)h(standard)0 2260 y(de\014nition)21 b([10])g(for)h(m)o(ultipro)q(cessor)e(Utilization)f (using)j Fk(p)g Fm(pro)q(cessors,)h Fk(U)5 b Fm(\()p Fk(p)p Fm(\))24 b(=)e Fk(E)s Fm(\()p Fk(p)p Fm(\))15 b Ff(\002)g Fk(R)p Fm(\()p Fk(p)p Fm(\),)0 2320 y(where)h Fk(R)p Fm(\()p Fk(p)p Fm(\))h(is)f(the)g(Redundancy)l(.)131 2447 y Fi(Utilization)43 b Fm(=)475 2374 y Fb( )551 2414 y Fi(p)n(ar)n(al)r(lelism\(new\))p 513 2436 425 2 v 513 2482 a(maximal)p 695 2482 15 2 v 19 w(p)n(ar)n(al)r (lelism)954 2447 y Ff(\002)1009 2414 y Fi(work\(max)p 1219 2414 V 18 w(p)n(ar\))p 1009 2436 314 2 v 1054 2482 a(work\(new\))1327 2374 y Fb(!)1371 2447 y Ff(\002)1471 2414 y Fi(work\(new\))p 1425 2436 V 1425 2482 a(work\(max)p 1635 2482 15 2 v 19 w(p)n(ar\))396 2585 y Fm(=)518 2551 y Fi(p)n(ar)n(al)r(lelism\(new\))p 480 2573 425 2 v 480 2619 a(maximal)p 662 2619 15 2 v 19 w(p)n(ar)n(al)r(lelism)0 2704 y Fm(where)16 b(w)o(e)g(note)g(that)h(Utilization)d(is)i(not)h(a)g (function)f(of)g(the)g(Redundancy)l(.)913 2828 y(12)p eop %%Page: 13 19 bop 0 195 a Fe(4.7)70 b(Amdahl's)22 b(La)n(w)0 287 y Fm(Amdahl's)f(La)o(w)j (is)e(a)i(sp)q(ecial)e(case)h(of)g(the)g(Sp)q(eedup)g(de\014nition)f(\(7\))i (in)e(whic)o(h)g(w)o(ork\(new\))h(=)0 348 y(w)o(ork\(old\),)642 435 y Fi(Sp)n(e)n(e)n(dup)14 b Fm(=)878 402 y Fi(p)n(ar)n(al)r(lelism\(new\)) p 878 424 350 2 v 888 469 a(p)n(ar)n(al)r(lelism\(old\))0 550 y Fm(where,)j(here,)f Fi(p)n(ar)n(al)r(lelism\(new\))i Fm(is)f(the)g (parallelism)e(in)i(a)g(theoretically)f(p)q(erfect)g(\(without)i(o)o(v)o(er-) 0 611 y(head\))e(optimized,)e(v)o(ectorized)g(or)j(concurren)o(tized)d(v)o (ersion)i(of)h(the)f(old)g(soft)o(w)o(are.)73 671 y(This)h(demonstrates)f (the)g(direct)f(relationship)i(b)q(et)o(w)o(een)e(parallelism,)f(p)q (erformance)h(and)i(tra-)0 731 y(ditional)h(p)q(erformance-related)g(metrics) e(and)j(leads)g(us)g(to)g(the)g(new)g(approac)o(h)g(to)g(p)q(erformance)0 791 y(ev)m(aluation)d(and)h(applications)f(b)q(enc)o(hmarking)f(that)i(is)f (describ)q(ed)g(in)f(the)h(next)g(section.)0 958 y Fl(5)83 b(Using)18 b(P)n(arallelism)g(to)g(Ev)-5 b(aluate)19 b(Application)f(Benc)n (h-)124 1049 y(marks)0 1158 y Fm(P)o(arallelism-based)g(tec)o(hniques)h(can)h (also)h(b)q(e)f(used)g(to)h(ev)m(aluate)f(application)f(b)q(enc)o(hmark)g Fi(sets)p Fm(.)0 1219 y(W)l(e)h(are)g(motiv)m(ated)e(to)j(study)f(the)f (parallelism)f(in)i(application)f(b)q(enc)o(hmark)g(sets)h(in)f(ligh)o(t)g (of)i(a)0 1279 y(common)15 b(metho)q(dology)h(no)o(w)i(b)q(eing)f(used)g(for) g(their)f(construction.)23 b(Often)17 b(b)q(enc)o(hmark)e(sets)i(are)0 1339 y(dev)o(elop)q(ed)c(b)o(y)h(selecting)g(programs)g(or)h(k)o(ernels)e (from)g(users)i(in)f(di\013eren)o(t)g(application)g(areas.)21 b(The)0 1399 y(application)i(areas)h(are)f(c)o(hosen)g(from)f(those)h(ha)o (ving)g(users)g(who)h(consume)e(large)h(amoun)o(ts)g(of)0 1459 y(sup)q(ercomputer)15 b(time)f([13].)21 b(Some)15 b(preliminary)e(results)j (in)g(other)g(studies)g([14][15])g(suggest)h(that)0 1520 y(while)10 b(the)h(a)o(v)o(erage)g(p)q(erformance)f(of)h(users)h(in)f(di\013eren)o(t)f (application)h(areas)h(is)f(somewhat)f(di\013eren)o(t,)0 1580 y(the)15 b(range)h(of)g(p)q(erformance)e(within)h(eac)o(h)g(area)h(is)f(so)h (large)g(as)g(to)g(mak)o(e)d(the)i(di\013erences)g(b)q(et)o(w)o(een)0 1640 y(them)g(insigni\014can)o(t.)23 b(It)17 b(is)g(not)g(curren)o(tly)f (clear)g(that)h(selecting)f(programs)h(for)h(application)f(area)0 1700 y(co)o(v)o(erage)f(alone)h(results)g(in)f(a)i(b)q(enc)o(hmark)d(set)i (whic)o(h)f(measurably)g(represen)o(ts)g(the)g(w)o(orkload)i(at)0 1760 y(a)e(particular)g(site)f(or)h(sup)q(ercomputer)f(usage)i(in)f(general.) k(A)c(more)f(rigorous)h(analytic)g(approac)o(h,)0 1821 y(p)q(erhaps)h(based)g (on)g(the)f(metho)q(d)f(describ)q(ed)h(in)g Ff(x)p Fm(2,)g(ma)o(y)f(b)q(e)h (more)f(appropriate.)73 1881 y(The)24 b(remainder)e(of)i(this)g(section)f (examines)f(the)i(executed)e(parallelism)g(of)i(some)f(curren)o(t)0 1941 y(b)q(enc)o(hmarks)f(on)i(the)f(CRA)l(Y)g(Y-MP)l(.)f(It)h(is)g(not)h(in) o(tended)f(to)g(b)q(e)h(an)g(exhaustiv)o(e)e(study)l(,)j(but)0 2001 y(rather)c(an)g(example)e(of)i(ho)o(w)h(some)e(simple)e (parallelism-based)i(tec)o(hniques)f(migh)o(t)g(b)q(e)j(used)f(to)0 2061 y(c)o(haracterize)g(the)i(executed)e(parallelism)g(of)i(application)g (programs)g(and)g(program)g(sets)g(on)h(a)0 2121 y(computer)15 b(system)g(ric)o(h)g(in)h(instruction-lev)o(el)e(parallelism.)20 b(While)15 b(the)h(to)q(ols)h(used)g(to)g(study)f(the)0 2182 y(Y-MP)f(are)h(mac)o(hine-sp)q(eci\014c,)d(the)i(parallelism-based)f(metho)q (dology)h(is)g(not,)h(and)g(w)o(e)f(hop)q(e)i(that)0 2242 y(other)f(researc)o (hers)g(will)f(apply)h(the)g(metho)q(dology)g(to)h(other)f(mac)o(hines.)73 2302 y(Our)11 b(approac)o(h)h(compleme)o(n)o(ts)c(the)j(w)o(ork)g(done)g(b)o (y)g(D.)f(K.)h(Chen,)h(H.)e(M.)g(Su)h(and)h(P)l(.)e(C.)h(Y)l(ew)g([16].)0 2362 y(They)20 b(ha)o(v)o(e)f(dev)o(elop)q(ed)g(a)i(to)q(ol,)g(MAXP)l(AR,)d (that)j(analyzes)f(the)g(soft)o(w)o(are)g(parallelism)e(that)i(is)0 2422 y(presen)o(t)13 b(in)g(F)l(ortran)h(source)g(co)q(de)g(at)f(the)h(job,)g (subroutine,)g(lo)q(op)g(and)g(statemen)o(t)e(lev)o(el,)f(assuming)0 2483 y(unlimited)h(hardw)o(are)j(resources.)21 b(Our)15 b(tec)o(hnique)e (measures)g(the)i(parallelism)d(that)j(w)o(as)h(actually)0 2543 y(exploited)f(on)j(the)e(\014nite)g(resources)h(of)g(the)f(CRA)l(Y)g (Y-MP)l(.)g(Measuremen)o(ts)e(tak)o(en)j(b)o(y)f(these)g(t)o(w)o(o)0 2603 y(tec)o(hniques)f(corresp)q(ond)i(to)f(the)g(placemen)o(t)e(of)i(t)o(w)o (o)g(prob)q(es)h(at)f(di\013eren)o(t)g(lo)q(cations)g(on)h(the)f(P)o(ath)0 2663 y(T)l(o)i(P)o(erformance)e(\()p Ff(x)q Fm(3\).)26 b(The)18 b(di\013erence)e(b)q(et)o(w)o(een)h(the)h(measuremen)o(ts)d(is)i(a)i(gauge)f (of)g(the)g(w)o(ork)913 2828 y(13)p eop %%Page: 14 20 bop 0 195 a Fm(done)17 b(b)o(y)e(the)h(agen)o(ts)h(who)g(acted)f(along)h(the) f(path)h(in)o(b)q(et)o(w)o(een)d(\(e.g.)21 b(compiler,)14 b(hardw)o(are\).)73 255 y(T)l(o)k(b)q(etter)g(understand)g(the)f(c)o(haracteristics)f(of)i (curren)o(t)f(w)o(orkloads,)h(in)f(another)h(study)g([14])0 315 y(w)o(e)d(are)g(also)g(examining)f(p)q(erformance)g(c)o(haracteristics)f (of)j(the)f(w)o(orkload)g(at)g(an)h(NSF)f(sup)q(ercom-)0 376 y(puter)i(cen)o(ter)e(o)o(v)o(er)h(a)h(one)g(y)o(ear)f(p)q(erio)q(d.)23 b(Com)o(bined)15 b(with)i(the)f(w)o(ork)h(describ)q(ed)f(here,)g(these)g(t)o (w)o(o)0 436 y(pro)s(jects)f(ma)o(y)g(enable)g(the)h(design)f(of)i(new)e(b)q (enc)o(hmark)g(sets)h(that)g(more)e(precisely)g(mo)q(del)h(actual)0 496 y(sup)q(ercomputer)f(usage.)21 b(Our)15 b(o)o(v)o(erall)e(goal)j(is)f(to) g(impro)o(v)o(e)d(the)i(tec)o(hniques)g(used)h(in)f(p)q(erformance)0 556 y(analysis)i(and,)h(in)f(particular,)f(b)q(enc)o(hmarking.)0 701 y Fe(5.1)70 b(Generating)22 b(P)n(arallelism)d(Matrices)j(for)i(the)e (CRA)-6 b(Y)22 b(Y-MP)0 793 y Fm(As)13 b(an)g(example)e(of)j(ho)o(w)f (parallelism-based)f(tec)o(hniques)f(migh)o(t)h(b)q(e)h(used)g(to)g(examine)e (application)0 853 y(programs)k(and)h(program)f(sets,)g(w)o(e)f(ha)o(v)o(e)h (measured)e(the)i(executed)f(parallelism)f(of)i(some)f(p)q(opular)0 913 y(b)q(enc)o(hmark)20 b(programs)i(using)g(a)g(clo)q(c)o(k)f(p)q(erio)q(d) h(lev)o(el)e(sim)o(ulator)g(of)i(a)g(single)f(pro)q(cessor)i(of)f(the)0 974 y(CRA)l(Y)14 b(Y-MP)h([12)q(].)20 b(The)15 b(sim)o(ulator)f(mo)q(dels)g (the)h(reserv)m(ation)g(of)h(the)f(follo)o(wing)g(CPU)g(resources)0 1034 y(of)d(the)f(CRA)l(Y)f(Y-MP:)h(all)g(v)o(ector)f(functional)h(units,)h (memory)c(p)q(orts,)13 b(v)o(ector)e(registers,)g(scalar)h(reg-)0 1094 y(isters.)22 b(Dela)o(ys)16 b(due)h(to)g(data-indep)q(enden)o(t)g(and)g (data-dep)q(enden)o(t)g(branc)o(hes,)f(instruction)h(bu\013er)0 1154 y(fetc)o(hes,)d(subroutine)i(calls,)f(instruction)g(issue,)h(and)g(some) f(memory)d(con\015icts)k(are)g(also)g(mo)q(deled.)0 1214 y(Scalar)j(input)f (path)h(con\015icts)f(and)h(some)e(t)o(yp)q(es)i(of)f(memory)e(con\015ict)i (dela)o(ys)g(are)g(not)h(mo)q(deled,)0 1275 y(but)i(in)g(most)f(cases)h(the)g (e\013ect)g(of)g(these)g(simpli\014cations)e(is)h(minimal;)g(except)g(for)h (malev)o(olen)o(t)0 1335 y(strides)16 b(and)h(gather/scatter)g(patterns,)f (the)g(sim)o(ulator)f(is)h(accurate)g(to)h(appro)o(ximately)d(5\045.)73 1395 y(One)k(of)f(the)h(textual)f(outputs)h(of)g(the)g(CRA)l(Y)e(Y-MP)i(sim)o (ulator)e(is)h(a)h(t)o(w)o(o-dimensional)e(par-)0 1455 y(allelism)e(matrix)g (of)j(the)f(t)o(yp)q(e)g(describ)q(ed)g(in)g Ff(x)p Fm(3.5.)22 b(The)16 b Fk(i;)8 b(j)s Fm(-th)16 b(en)o(try)g(of)h(the)f(parallelism)e (matrix)0 1515 y(represen)o(ts)20 b(the)h(fraction)g(of)g(the)g(program)f (elapsed)h(time)e(when)i Fk(i)f Fm(memory)e(op)q(erations)23 b(and)e Fk(j)0 1576 y Fm(\015oating)e(p)q(oin)o(t)f(op)q(erations)h(w)o(ere)e (sim)o(ultaneously)f(completed.)24 b(The)18 b(Y-MP)f(has)i(three)e (\015oating)0 1636 y(p)q(oin)o(t)g(functional)g(units)h(\(add,)f(m)o(ultiply) l(,)d(and)k(recipro)q(cal)e(appro)o(ximation\))g(and)i(three)f(memory)0 1696 y(p)q(orts)e(\(t)o(w)o(o)f(load)g(and)h(one)f(store\),)g(consequen)o (tly)f(the)g(parallelism)f(matrices)g(used)i(in)g(this)f(section)0 1756 y(are)j(4)c Ff(\002)f Fm(4.)73 1816 y(The)17 b(sim)o(ulator)e(w)o(as)i (applied)f(to)h(a)g(subset)g(of)g(the)f(P)o(erfect)g(Benc)o(hmarks)e([3])i (to)h(measure)e(the)0 1876 y(executed)j(parallelism)f(of)j(sev)o(eral)e (application)h(programs)g(in)g(the)g(set.)31 b(Fiv)o(e)17 b(of)j(the)f(13)h (P)o(erfect)0 1937 y(Benc)o(hmarks)9 b(w)o(ere)h(sim)o(ulated.)18 b(The)11 b(execution)f(time)f(required)g(to)j(sim)o(ulate)d(eac)o(h)h(en)o (tire)g(program)0 2159 y Fn(T)-5 b(able)14 b(1)p Fm(:)20 b(Names,)11 b(application)i(areas,)h(and)f(sim)o(ulation)e(statistics)h(of)h(the)g(\014v) o(e)f(sim)o(ulated)e(P)o(erfect)0 2220 y(Benc)o(hmarks)1030 2266 y(\045)16 b(of)h(Program)49 b(Sim)o(ulation)269 2326 y(Program)74 b(Application)15 b(Area)179 b(Sim)o(ulated)147 b(Error)p 244 2346 1388 2 v 269 2388 a(ADM)140 b(Meteorology)312 b(98.85\045)166 b(-6.43\045)269 2448 y(AR)o(C2D)91 b(Aero)q(dynamics)271 b(97.46\045)166 b(-6.71\045)269 2508 y(BDNA)113 b(Ph)o(ysical)15 b(Chemistry)159 b(72.65\045)166 b(-3.41\045)269 2569 y(D)o(YFESM)49 b(Structural)16 b(Mec)o(hanics)121 b(99.75\045)166 b(-6.25\045)269 2629 y(MDG)139 b(Ph)o(ysical)15 b(Chemistry)159 b(90.32\045)166 b(-2.27\045)913 2828 y(14)p eop %%Page: 15 21 bop 465 196 a Fm(FLOPS)p 272 215 547 2 v 223 256 a(3)p 271 274 2 61 v 50 w(0.00)50 b(0.00)h(0.00)f(0.00)p 817 274 V 51 w(M)223 316 y(2)p 271 334 V 50 w(0.02)g(0.00)h(0.00)f(0.00)p 817 334 V 54 w(O)223 376 y(1)p 271 394 V 50 w(0.12)g(0.01)h(0.00)f(0.00)p 817 394 V 57 w(P)223 436 y(0)p 271 455 V 50 w(0.75)g(0.09)h(0.01)f(0.00)p 817 455 V 60 w(S)p 272 456 547 2 v 359 497 a(0)113 b(1)f(2)h(3)476 557 y Fn(ADM)1229 196 y Fm(FLOPS)p 1037 215 V 987 256 a(3)p 1036 274 2 61 v 51 w(0.01)50 b(0.03)h(0.03)f(0.00)p 1582 274 V 51 w(M)987 316 y(2)p 1036 334 V 51 w(0.05)g(0.13)h(0.08)f(0.00)p 1582 334 V 54 w(O)987 376 y(1)p 1036 394 V 51 w(0.07)g(0.20)h(0.11)f(0.00)p 1582 394 V 56 w(P)987 436 y(0)p 1036 455 V 51 w(0.07)g(0.13)h(0.07)f(0.02)p 1582 455 V 59 w(S)p 1037 456 547 2 v 1124 497 a(0)112 b(1)h(2)f(3)1213 557 y Fn(AR)n(C2D)465 677 y Fm(FLOPS)p 272 697 V 223 737 a(3)p 271 755 2 61 v 50 w(0.00)50 b(0.02)h(0.02)f(0.00)p 817 755 V 51 w(M)223 798 y(2)p 271 816 V 50 w(0.01)g(0.05)h(0.04)f(0.00)p 817 816 V 54 w(O)223 858 y(1)p 271 876 V 50 w(0.05)g(0.12)h(0.07)f(0.00)p 817 876 V 57 w(P)223 918 y(0)p 271 936 V 50 w(0.28)g(0.17)h(0.11)f(0.03)p 817 936 V 60 w(S)p 272 938 547 2 v 359 978 a(0)113 b(1)f(2)h(3)461 1038 y Fn(BDNA)1229 677 y Fm(FLOPS)p 1037 697 V 987 737 a(3)p 1036 755 2 61 v 51 w(0.00)50 b(0.00)h(0.01)f(0.00)p 1582 755 V 51 w(M)987 798 y(2)p 1036 816 V 51 w(0.03)g(0.02)h(0.02)f(0.00)p 1582 816 V 54 w(O)987 858 y(1)p 1036 876 V 51 w(0.12)g(0.02)h(0.02)f(0.00)p 1582 876 V 56 w(P)987 918 y(0)p 1036 936 V 51 w(0.68)g(0.08)h(0.01)f(0.00)p 1582 936 V 59 w(S)p 1037 938 547 2 v 1124 978 a(0)112 b(1)h(2)f(3)1190 1038 y Fn(D)n(YFESM)847 1163 y Fm(FLOPS)p 654 1183 V 605 1223 a(3)p 653 1241 2 61 v 50 w(0.00)51 b(0.00)f(0.00)h(0.00)p 1200 1241 V 50 w(M)605 1283 y(2)p 653 1301 V 50 w(0.01)g(0.00)f(0.00)h(0.00)p 1200 1301 V 54 w(O)605 1344 y(1)p 653 1362 V 50 w(0.10)g(0.00)f(0.00)h(0.00)p 1200 1362 V 56 w(P)605 1404 y(0)p 653 1422 V 50 w(0.77)g(0.12)f(0.00)h(0.00)p 1200 1422 V 59 w(S)p 654 1423 547 2 v 742 1464 a(0)112 b(1)h(2)f(3)857 1524 y Fn(MDG)263 1676 y(T)-5 b(able)19 b(2)p Fm(:)i(P)o(arallelism)13 b(Matrices)j(for)g(selected)f(P)o(erfect)g(Benc)o(hmarks)0 1801 y(is)f(prohibitiv)o(e,)f(so)i(w)o(e)g(sim)o(ulated)d(only)j(a)g(subset)f (of)h(the)g(calling)f(tree)f(of)i(the)g(program.)20 b(Ho)o(w)o(ev)o(er,)0 1861 y(in)c(eac)o(h)g(case,)f(the)h(subtree)g(is)g(called)f(from)h(a)g(DO)h (lo)q(op)g(con)o(tained)e(in)h(the)g(paren)o(t)g(subroutine)h(of)0 1921 y(the)i(subtree's)f(ro)q(ot.)31 b(This)20 b(means)e(that)h(the)g (subtree)g(is)g(called)f(rep)q(eatedly)g(and)i(w)o(e)f(need)f(only)0 1981 y(sim)o(ulate)12 b(one)i(instan)o(tiation)g(to)h(capture)f(the)f (desired)h(measuremen)o(ts.)j(Eac)o(h)d(subtree)g(accoun)o(ted)0 2042 y(for)21 b(b)q(et)o(w)o(een)e(72\045)i(and)g(99\045)g(of)f(the)h(total)f (execution)f(time)g(of)h(the)h(whole)f(program.)33 b(T)l(able)21 b(1)0 2102 y(lists)c(the)g(b)q(enc)o(hmark)f(programs)i(that)g(w)o(ere)f(sim) o(ulated)e(and)j(the)g(application)f(areas)h(that)g(these)0 2162 y(b)q(enc)o(hmark)d(programs)h(represen)o(t.)0 2306 y Fe(5.2)70 b(P)n(arallelism)19 b(Matrices)j(for)h(Selected)e(P)n(erfect)g (Benc)n(hmarks)73 2399 y Fm(T)l(able)f(2)g(sho)o(ws)g(the)g(parallelism)d (matrices)h(for)i(the)f(sim)o(ulated)e(programs.)32 b(Man)o(y)19 b(features)0 2459 y(of)h(program)g(b)q(eha)o(vior)f(can)h(b)q(e)g(observ)o (ed)f(directly)f(and)i(quan)o(ti\014ed)f(from)g(the)g(parallelism)f(ma-)0 2519 y(trix.)34 b(A)20 b(c)o(hec)o(k)f(of)i(elemen)o(t)d(\(0)p Fk(;)8 b Fm(0\))21 b(imme)o(diately)c(rev)o(eals)j(whether)g(the)g(program)h (is)f(scalar-)i(or)0 2579 y(v)o(ector-orien)o(ted.)34 b(Larger)22 b(v)m(alues)f(indicate)f(that)i(neither)e(\015oating-p)q(oin)o(t)i(nor)g (memory)c(op)q(era-)0 2640 y(tions)i(o)q(ccurred)g(during)g(most)g(of)g(the)g (elapsed)g(time,)e(suc)o(h)i(as)h(w)o(ould)f(b)q(e)g(the)g(case)g(in)f(a)i (scalar)0 2700 y(co)q(de.)29 b(W)l(e)18 b(also)i(note)e(the)h(v)o(ery)e(lo)o (w)i(fraction)g(of)g(elapsed)f(time)f(in)h(b)q(oth)i(column)d(3)i(and)g(ro)o (w)g(3,)913 2828 y(15)p eop %%Page: 16 22 bop 203 514 2 360 v 188 515 16 2 v 188 442 V 188 370 V 188 298 V 188 226 V 188 154 V 188 478 V 188 406 V 188 334 V 188 262 V 188 190 V 84 528 a Fa(0.0)84 456 y(0.2)84 384 y(0.4)84 312 y(0.6)84 240 y(0.8)84 168 y(1.0)p 203 515 322 2 v 234 538 2 25 v 483 538 V 275 529 2 16 v 317 529 V 358 529 V 400 529 V 441 529 V 483 529 V 223 582 a(0)226 b(6)321 630 y(adm)p 226 514 17 270 v 268 514 17 76 v 309 514 17 15 v 351 514 17 4 v 508 514 2 360 v 494 515 16 2 v 494 478 V 494 442 V 494 406 V 494 370 V 494 334 V 494 298 V 494 262 V 494 226 V 494 190 V 494 154 V 509 515 322 2 v 539 538 2 25 v 789 538 V 581 529 2 16 v 623 529 V 664 529 V 706 529 V 747 529 V 789 529 V 529 582 a(0)g(6)616 630 y(arc2d)p 532 514 17 26 v 574 514 17 72 v 615 514 17 112 v 657 514 17 98 v 698 514 17 44 v 740 514 17 11 v 814 514 2 360 v 800 515 16 2 v 800 478 V 800 442 V 800 406 V 800 370 V 800 334 V 800 298 V 800 262 V 800 226 V 800 190 V 800 154 V 815 515 322 2 v 845 538 2 25 v 1094 538 V 887 529 2 16 v 928 529 V 970 529 V 1011 529 V 1053 529 V 1094 529 V 835 582 a(0)g(6)926 630 y(b)q(dna)p 838 514 17 101 v 879 514 17 83 v 921 514 17 87 v 962 514 17 58 v 1004 514 17 22 v 1045 514 17 8 v 1120 514 2 360 v 1106 515 16 2 v 1106 478 V 1106 442 V 1106 406 V 1106 370 V 1106 334 V 1106 298 V 1106 262 V 1106 226 V 1106 190 V 1106 154 V 1121 515 322 2 v 1151 538 2 25 v 1400 538 V 1192 529 2 16 v 1234 529 V 1275 529 V 1317 529 V 1358 529 V 1400 529 V 1140 582 a(0)g(6)1212 630 y(dyfesm)p 1143 514 17 245 v 1185 514 17 69 v 1226 514 17 22 v 1268 514 17 15 v 1310 514 17 8 v 1351 514 17 4 v 1426 514 2 360 v 1411 515 16 2 v 1411 478 V 1411 442 V 1411 406 V 1411 370 V 1411 334 V 1411 298 V 1411 262 V 1411 226 V 1411 190 V 1411 154 V 1426 515 322 2 v 1457 538 2 25 v 1706 538 V 1498 529 2 16 v 1540 529 V 1581 529 V 1623 529 V 1664 529 V 1706 529 V 1446 582 a(0)g(6)1544 630 y(mdg)p 1449 514 17 278 v 1491 514 17 80 v 1532 514 17 4 v 269 515 1204 2 v 497 754 2 57 v 581 737 a(0)107 b(1)h(2)f(3)h(4)g(5)f(6)p 1411 754 V 50 w(Av)o(erage)p 253 755 1370 2 v 278 795 a(ADM)p 497 812 2 57 v 134 w(0.75)48 b(0.21)h(0.04)f(0.01)h(0.00)g(0.00)f(0.00)p 1411 812 V 77 w(0.305)278 851 y(AR)o(C2D)p 497 868 V 88 w(0.07)g(0.20)h(0.31) f(0.27)h(0.12)g(0.03)f(0.00)p 1411 868 V 77 w(2.278)278 908 y(BDNA)p 497 925 V 110 w(0.28)g(0.23)h(0.24)f(0.16)h(0.06)g(0.02)f(0.00)p 1411 925 V 77 w(1.566)278 964 y(D)o(YFESM)p 497 981 V 49 w(0.68)g(0.19)h (0.06)f(0.04)h(0.02)g(0.01)f(0.00)p 1411 981 V 77 w(0.534)278 1021 y(MDG)p 497 1038 V 132 w(0.77)g(0.22)h(0.01)f(0.00)h(0.00)g(0.00)f(0.00) p 1411 1038 V 77 w(0.244)198 1172 y Fn(Figure)18 b(2)p Fm(:)j(T)l(otal)c (parallelism)d(v)o(ectors)h(for)i(selected)d(P)o(erfect)h(Benc)o(hmarks)0 1306 y(indicating)g(none)h(of)g(these)f(co)q(des)i(highly)e(utilize)f(all)h (3)h(arithmetic)d(units)j(or)g(all)f(3)h(memory)d(p)q(orts)0 1366 y(at)k(the)f(same)f(time.)73 1426 y(Figure)d(2)i(sho)o(ws)f(the)g (parallelism)d(v)o(ectors)i(of)h(the)g(5)g(b)q(enc)o(hmarks)e(where)i(\\w)o (ork")g(is)g(de\014ned)f(as)0 1487 y(either)k(a)h(\015oating-p)q(oin)o(t)h (or)f(memory)d(op)q(eration.)24 b(Because)16 b(in)h(this)f(case)h(w)o(e)g(do) g(not)g(distinguish)0 1547 y(b)q(et)o(w)o(een)e(the)h(three)f(\015oating)i(p) q(oin)o(t)f(functional)g(units)g(and)h(three)e(memory)e(p)q(orts)k(on)g(the)f (Y-MP)l(,)0 1607 y(the)d(parallelism)d(v)o(ectors)i(con)o(tain)h(en)o(tries)f (for)h(0)g(to)h(6)f(concurren)o(tly)e(completed)g(op)q(erations.)21 b(Note)0 1667 y(that)14 b(the)f(parallelism)e(reac)o(hes)h(a)i(lev)o(el)d(of) j(5)f(in)g(a)h(small)d(fraction)j(of)f(the)g(normalized)f(time)f(in)i(a)g (few)0 1727 y(of)i(the)g(b)q(enc)o(hmarks.)20 b(In)15 b(this)g(and)g(the)g (other)g(parallelism)e(v)o(ectors)i(describ)q(ed)f(later,)h(the)f(a)o(v)o (erage)0 1787 y(parallelism)c(is)j(the)f(w)o(eigh)o(ted)g(sum)f(of)i(the)g (normalized)d(time)h Fk(p)1178 1794 y Fc(i)1205 1787 y Fm(sp)q(en)o(t)i(at)g (eac)o(h)f(parallelism)e(lev)o(el)h Fk(i)p Fm(.)606 1924 y Fi(A)o(ver)n(age)18 b(p)n(ar)n(al)r(lelism)d Fm(=)1104 1870 y Fc(n)1084 1882 y Fb(X)1086 1973 y Fc(i)p Fj(=0)1153 1924 y Fk(i)10 b Ff(\002)h Fk(p)1254 1931 y Fc(i)0 2066 y Fm(Figure)i(3)h(sho)o (ws)h(parallelism)c(v)o(ectors)i(and)h(a)o(v)o(erage)g(parallelism)d(for)j (the)f(\014v)o(e)g(b)q(enc)o(hmarks)f(where)0 2126 y(\\w)o(ork")j(is)g (de\014ned)f(as)h(only)g(\015oating)g(p)q(oin)o(t)g(op)q(erations.)21 b(Similarly)l(,)12 b(Figure)i(4)h(sho)o(ws)g(parallelism)0 2186 y(data)i(where)f(w)o(ork)g(is)g(de\014ned)g(as)h(only)f(memory)e(op)q (erations.)0 2331 y Fe(5.3)70 b(Di\013erences)26 b(in)j(Executed)f(P)n (arallelism)d(for)30 b(Selected)c(P)n(erfect)157 2405 y(Benc)n(hmarks)73 2498 y Fm(The)14 b(di\013erence)e(in)h(parallelism)e(in)i(the)g(5)h(P)o (erfect)e(Benc)o(hmarks)f(can)i(b)q(e)h(examined)d(visually)h(in)0 2558 y(Figures)j(2)g(through)h(4,)g(or)f(quan)o(titativ)o(ely)e(using)i(the)g (distance)g(metric)e(in)h(equation)h(\(1\).)22 b(T)l(able)15 b(3)0 2618 y(sho)o(ws)23 b(the)g(quan)o(titativ)o(e)d(di\013erence)i(in)g (parallelism)e(for)j(eac)o(h)f(pair)g(of)h(the)f(\014v)o(e)g(b)q(enc)o (hmarks.)0 2678 y(Note)g(that)h(the)f(di\013erence)f(in)h(parallelism)e(is)i (not)h(a)f(transitiv)o(e)g(relation.)39 b(W)l(e)22 b(\014rst)g(compare)913 2828 y(16)p eop %%Page: 17 23 bop 410 514 2 360 v 396 515 16 2 v 396 442 V 396 370 V 396 298 V 396 226 V 396 154 V 396 478 V 396 406 V 396 334 V 396 262 V 396 190 V 292 528 a Fa(0.0)292 456 y(0.2)292 384 y(0.4)292 312 y(0.6)292 240 y(0.8)292 168 y(1.0)p 411 515 239 2 v 441 538 2 25 v 607 538 V 483 529 2 16 v 524 529 V 566 529 V 607 529 V 431 582 a(0)143 b(4)487 630 y(adm)p 434 514 17 307 v 475 514 17 47 v 517 514 17 8 v 633 514 2 360 v 619 515 16 2 v 619 478 V 619 442 V 619 406 V 619 370 V 619 334 V 619 298 V 619 262 V 619 226 V 619 190 V 619 154 V 634 515 239 2 v 664 538 2 25 v 830 538 V 706 529 2 16 v 747 529 V 789 529 V 830 529 V 653 582 a(0)h(4)699 630 y(arc2d)p 657 514 17 105 v 698 514 17 137 v 740 514 17 94 v 781 514 17 29 v 856 514 2 360 v 841 515 16 2 v 841 478 V 841 442 V 841 406 V 841 370 V 841 334 V 841 298 V 841 262 V 841 226 V 841 190 V 841 154 V 856 515 239 2 v 887 538 2 25 v 1053 538 V 928 529 2 16 v 970 529 V 1011 529 V 1053 529 V 876 582 a(0)f(4)926 630 y(b)q(dna)p 879 514 17 127 v 921 514 17 134 v 962 514 17 87 v 1004 514 17 11 v 1078 514 2 360 v 1064 515 16 2 v 1064 478 V 1064 442 V 1064 406 V 1064 370 V 1064 334 V 1064 298 V 1064 262 V 1064 226 V 1064 190 V 1064 154 V 1079 515 239 2 v 1109 538 2 25 v 1275 538 V 1151 529 2 16 v 1192 529 V 1234 529 V 1275 529 V 1099 582 a(0)g(4)1129 630 y(dyfesm)p 1102 514 17 299 v 1143 514 17 44 v 1185 514 17 19 v 1301 514 2 360 v 1287 515 16 2 v 1287 478 V 1287 442 V 1287 406 V 1287 370 V 1287 334 V 1287 298 V 1287 262 V 1287 226 V 1287 190 V 1287 154 V 1302 515 239 2 v 1332 538 2 25 v 1498 538 V 1374 529 2 16 v 1415 529 V 1457 529 V 1498 529 V 1322 582 a(0)g(4)1378 630 y(mdg)p 1325 514 17 321 v 1366 514 17 37 v 1408 514 17 4 v 476 515 872 2 v 682 757 2 61 v 770 739 a Fm(0)113 b(1)f(2)h(3)p 1228 757 V 50 w(Av)o(erage)p 424 759 1028 2 v 449 801 a(ADM)p 682 819 2 61 v 140 w(0.88)50 b(0.11)h(0.01)g(0.00)p 1228 819 V 81 w(0.127)449 861 y(AR)o(C2D)p 682 879 V 91 w(0.19)f(0.49)h(0.30)g(0.02)p 1228 879 V 81 w(1.149)449 921 y(BDNA)p 682 940 V 113 w(0.35)f(0.37)h(0.24)g (0.03)p 1228 940 V 81 w(0.956)449 982 y(D)o(YFESM)p 682 1000 V 49 w(0.83)f(0.12)h(0.05)g(0.00)p 1228 1000 V 81 w(0.223)449 1042 y(MDG)p 682 1060 V 139 w(0.88)f(0.12)h(0.00)g(0.00)p 1228 1060 V 81 w(0.125)109 1194 y Fn(Figure)18 b(3)p Fm(:)k(Floating)16 b(p)q(oin)o(t)g(parallelism)e(v)o(ectors)i(for)g(selected)f(P)o(erfect)g (Benc)o(hmarks)0 1328 y(BDNA)d(and)i(MDG,)f(t)o(w)o(o)g(programs)g(represen)o (ting)f(the)h(application)g(area)h(of)f(Ph)o(ysical)f(Chemistry)l(.)0 1389 y(The)17 b(relativ)o(ely)d(high)i(v)m(alue,)g(0.53,)h(of)g(the)f (di\013erence)g(in)g(parallelism)e(illustrates)i(that)h(these)f(t)o(w)o(o)0 1449 y(programs)d(ha)o(v)o(e)f(di\013eren)o(t)g(parallelism)e(pro\014les.)20 b(Our)13 b(measuremen)o(ts)d(tell)h(us)j(that)f(eac)o(h)f(program)0 1509 y(exercises)j(the)h(CRA)l(Y)g(Y-MP)g(CPU)h(with)f(a)h(di\013eren)o(t)f (mix)f(of)i(parallelism.)i(Our)e(measuremen)o(ts)0 1569 y(do)c(not)h(tell)e (us)h(is)g(ho)o(w)g(w)o(ell)f(these)g(t)o(w)o(o)h(programs)g(represen)o(t)f (the)h(parallelism)e(pro\014les)i(of)g(Ph)o(ysical)0 1629 y(Chemistry)h (application)i(programs)h(in)f(general.)73 1690 y(Next)c(compare)g(ADM,)g(a)h (program)g(represen)o(ting)f(the)g(application)h(area)g(of)h(Meteorology)l(,) e(and)0 1750 y(MDG,)f(a)g(Ph)o(ysical)g(Chemistry)e(co)q(de.)19 b(The)12 b(relativ)o(ely)c(lo)o(w)j(v)m(alue)g(of)g(the)g(di\013erence)f(in)h (parallelism,)0 1810 y(0.04,)k(illustrates)f(that)h(these)f(t)o(w)o(o)h (programs)f(ha)o(v)o(e)g(v)o(ery)g(similar)e(parallelism)g(pro\014les.)21 b(Although)0 1870 y(the)j(t)o(w)o(o)g(programs)g(come)f(from)g(di\013eren)o (t)g(application)h(areas,)j(eac)o(h)d(program)g(exercises)e(the)0 1930 y(CRA)l(Y)c(Y-MP)h(CPU)h(with)f(a)h(v)o(ery)e(similar)f(mix)g(of)j (parallelism.)27 b(Our)20 b(measuremen)o(t)o(s)d(suggest)0 1990 y(that)g(the)f(presence)g(of)h(ADM)f(and)i(MDG)e(in)h(the)f(P)o(erfect)g (Benc)o(hmarks)e(set)i(ma)o(y)g(b)q(e)g(redundan)o(t,)0 2051 y(in)i(so)h(far)g(as)g(their)e(suitabilit)o(y)g(for)i(testing)f(the)g(CRA)l (Y.)f(One)h(migh)o(t)f(susp)q(ect)h(that)h(they)f(w)o(ould)p 561 2204 V 587 2186 a(ADM)49 b(AR)o(C2D)h(BDNA)f(D)o(YFESM)g(MDG)p 303 2205 1270 2 v 328 2247 a(ADM)p 561 2265 2 61 v 156 w(0.00)328 2308 y(AR)o(C2D)p 561 2326 V 107 w(0.74)107 b(0.00)328 2368 y(BDNA)p 561 2386 V 129 w(0.51)g(0.26)120 b(0.00)328 2428 y(D)o(YFESM)p 561 2446 V 65 w(0.08)107 b(0.66)120 b(0.45)141 b(0.00)328 2488 y(MDG)p 561 2506 V 155 w(0.04)107 b(0.76)120 b(0.53)141 b(0.11)129 b(0.00)296 2643 y Fn(T)-5 b(able)18 b(3)p Fm(:)j(Di\013erences)16 b(in)g(P)o(arallelism)d(\(F)l(rob)q(enius)k(error)f(norms\))913 2828 y(17)p eop %%Page: 18 24 bop 410 514 2 360 v 396 515 16 2 v 396 442 V 396 370 V 396 298 V 396 226 V 396 154 V 396 478 V 396 406 V 396 334 V 396 262 V 396 190 V 292 528 a Fa(0.0)292 456 y(0.2)292 384 y(0.4)292 312 y(0.6)292 240 y(0.8)292 168 y(1.0)p 411 515 239 2 v 441 538 2 25 v 607 538 V 483 529 2 16 v 524 529 V 566 529 V 607 529 V 431 582 a(0)143 b(4)487 630 y(adm)p 434 514 17 307 v 475 514 17 47 v 517 514 17 8 v 633 514 2 360 v 619 515 16 2 v 619 478 V 619 442 V 619 406 V 619 370 V 619 334 V 619 298 V 619 262 V 619 226 V 619 190 V 619 154 V 634 515 239 2 v 664 538 2 25 v 830 538 V 706 529 2 16 v 747 529 V 789 529 V 830 529 V 653 582 a(0)h(4)699 630 y(arc2d)p 657 514 17 105 v 698 514 17 137 v 740 514 17 94 v 781 514 17 29 v 856 514 2 360 v 841 515 16 2 v 841 478 V 841 442 V 841 406 V 841 370 V 841 334 V 841 298 V 841 262 V 841 226 V 841 190 V 841 154 V 856 515 239 2 v 887 538 2 25 v 1053 538 V 928 529 2 16 v 970 529 V 1011 529 V 1053 529 V 876 582 a(0)f(4)926 630 y(b)q(dna)p 879 514 17 217 v 921 514 17 87 v 962 514 17 37 v 1004 514 17 19 v 1078 514 2 360 v 1064 515 16 2 v 1064 478 V 1064 442 V 1064 406 V 1064 370 V 1064 334 V 1064 298 V 1064 262 V 1064 226 V 1064 190 V 1064 154 V 1079 515 239 2 v 1109 538 2 25 v 1275 538 V 1151 529 2 16 v 1192 529 V 1234 529 V 1275 529 V 1099 582 a(0)g(4)1129 630 y(dyfesm)p 1102 514 17 278 v 1143 514 17 54 v 1185 514 17 26 v 1226 514 17 4 v 1301 514 2 360 v 1287 515 16 2 v 1287 478 V 1287 442 V 1287 406 V 1287 370 V 1287 334 V 1287 298 V 1287 262 V 1287 226 V 1287 190 V 1287 154 V 1302 515 239 2 v 1332 538 2 25 v 1498 538 V 1374 529 2 16 v 1415 529 V 1457 529 V 1498 529 V 1322 582 a(0)g(4)1378 630 y(mdg)p 1325 514 17 321 v 1366 514 17 37 v 1408 514 17 4 v 476 515 872 2 v 682 757 2 61 v 770 739 a Fm(0)113 b(1)f(2)h(3)p 1228 757 V 50 w(Av)o(erage)p 424 759 1028 2 v 449 801 a(ADM)p 682 819 2 61 v 140 w(0.85)50 b(0.13)h(0.02)g(0.00)p 1228 819 V 81 w(0.178)449 861 y(AR)o(C2D)p 682 879 V 91 w(0.29)f(0.38)h(0.26)g(0.08)p 1228 879 V 81 w(1.128)449 921 y(BDNA)p 682 940 V 113 w(0.60)f(0.24)h(0.10)g (0.05)p 1228 940 V 81 w(0.611)449 982 y(D)o(YFESM)p 682 1000 V 49 w(0.77)f(0.15)h(0.07)g(0.01)p 1228 1000 V 81 w(0.311)449 1042 y(MDG)p 682 1060 V 139 w(0.89)f(0.10)h(0.01)g(0.00)p 1228 1060 V 81 w(0.118)102 1194 y Fn(Figure)18 b(4)p Fm(:)j(Memory)15 b(access)h(parallelism)e(v)o(ectors)h(for)i(selected)d(P)o(erfect)h(Benc)o (hmarks)0 1328 y(also)23 b(b)q(e)f(redundan)o(t)g(for)h(testing)f(other)g (mac)o(hines)e(with)i(v)o(ector)f(arc)o(hitectures)g(similar)f(to)i(the)0 1389 y(CRA)l(Y,)12 b(but)i(this)g(h)o(yp)q(othesis)f(needs)h(to)g(b)q(e)g(in) o(v)o(estigated.)19 b(The)13 b(same)g(conclusion)g(of)h(redundancy)0 1449 y(migh)o(t)21 b(b)q(e)h(also)h(b)q(e)g(dra)o(wn)g(from)e(our)i (measuremen)o(t)o(,)e(0.08,)j(of)f(the)f(di\013erence)f(in)h(parallelism)0 1509 y(in)17 b(D)o(YFESM,)f(a)h(program)g(represen)o(ting)f(the)h (application)g(area)h(of)f(Structural)g(Mec)o(hanics,)e(and)0 1569 y(MDG.)0 1714 y Fe(5.4)70 b(Prob)r(e)23 b(P)n(oin)n(ts)f(for)i(the)e (Baseline)f(P)n(erfect)h(Benc)n(hmarks)0 1806 y Fm(It)17 b(is)g(imp)q(ortan)o (t)g(to)h(realize)e(that)h(these)h(results)f(are)g(for)h(the)f(\\baseline")h (v)o(ersion)e(of)i(the)f(P)o(erfect)0 1866 y(Benc)o(hmarks.)26 b(In)19 b(the)f(P)o(erfect)g(metho)q(dology)l(,)g(baseline)g(refers)g(to)h(b) q(enc)o(hmark)e(exp)q(erimen)o(ts)f(in)0 1926 y(whic)o(h)c(no)i(man)o(ual)e (mo)q(di\014cations)h(can)g(b)q(e)g(made)f(to)i(the)f(source)g(co)q(de.)20 b(In)13 b(the)g(terminology)e(of)j Ff(x)p Fm(3,)0 1986 y(this)i(corresp)q (onds)h(to)g(the)f(placemen)o(t)d(of)k(a)f(prob)q(e)h(on)f(the)g(P)o(ath)h (to)f(P)o(erformance)e(at)j(the)f(lev)o(el)e(of)0 2047 y(\\Program.")22 b(The)16 b(P)o(erfect)f(metho)q(dology)g(allo)o(ws)i(for)f(another)h(t)o(yp)q (e)e(of)h(b)q(enc)o(hmark)f(exp)q(erimen)o(t)0 2107 y(called)h(\\optimized")f (in)h(whic)o(h)g(an)o(y)h(man)o(ual)e(mo)q(di\014cations)h(to)h(the)g (program)f(are)h(allo)o(w)o(ed.)22 b(The)0 2167 y(only)15 b(restriction)f(is) h(that)h(the)e(correct)h(solution)g(to)h(the)f(same)f(problem)f(b)q(e)j (obtained.)21 b(This)15 b(other)0 2227 y(b)q(enc)o(hmark)k(exp)q(erimen)o(t)f (corresp)q(onds)23 b(to)e(a)g(placemen)o(t)e(of)i(a)h(prob)q(e)f(on)h(the)e (P)o(ath)i(to)f(P)o(erfor-)0 2287 y(mance)d(at)i(the)f(lev)o(el)e(of)j (\\Algorithm")e(\\Mathematical)g(Mo)q(del,")i(or)f(\\Ph)o(ysical)g (Phenomenon")0 2348 y(dep)q(ending)c(on)g(what)h(represen)o(tation)e(of)i (the)e(problem)g(is)g(constan)o(t)i(and)f(the)g(exten)o(t)f(of)h(the)g(man-)0 2408 y(ual)i(mo)q(di\014cations)g(to)g(the)g(program.)24 b(Ev)o(en)16 b(though)i(the)f(results)g(presen)o(ted)f(here)h(do)h(not)f(apply)0 2468 y(to)g(this)g(other)g(exp)q(erimen)o(t,)c(it)j(w)o(ould)h(b)q(e)g(in)o (teresting)f(to)h(examine)d(the)j(parallelism)d(of)j(the)g(opti-)0 2528 y(mized)f(co)q(des,)j(as)g(it)f(w)o(ould)h(pro)o(vide)e(insigh)o(t)h(in) o(to)g(the)g(amoun)o(t)g(of)h(lost)f(parallelism)e(that)j(could)0 2588 y(ha)o(v)o(e)e(b)q(een)h(preserv)o(ed)e(when)i(other)g(agen)o(ts)h (\(e.g.)25 b(Soft)o(w)o(are)18 b(Engineer,)f(Numerical)e(Analyst,)j(or)0 2649 y(Scien)o(tist\))d(activ)o(ely)f(participate)h(to)i(preserv)o(e)e(the)h (parallelism.)913 2828 y(18)p eop %%Page: 19 25 bop 0 203 a Fl(6)83 b(Conclusions)25 b(and)j(F)-7 b(uture)27 b(W)-7 b(ork)0 313 y Fm(In)13 b(this)g(pap)q(er)g(w)o(e)g(presen)o(ted)f(a)h (brief)g(o)o(v)o(erview)e(of)i(p)q(erformance)f(ev)m(aluation)h(and)h(b)q (enc)o(hmarking.)0 373 y(W)l(e)j(demonstrated)g(that)i(traditional)e(p)q (erformance)g(measuremen)o(ts)d(in)k(these)f(activities)f(are)i(re-)0 433 y(ally)c(a)i(direct)d(measuremen)o(t)f(of)j(the)g(parallelism)e(in)h(the) h(soft)o(w)o(are)g(and)h(hardw)o(are.)21 b(A)14 b(framew)o(ork)0 493 y(called)i(the)g(P)o(ath)h(to)h(P)o(erformance)d(w)o(as)i(dev)o(elop)q (ed)f(whic)o(h)g(iden)o(ti\014es)f(the)i(agen)o(ts)g(and)h(activities)0 554 y(that)g(c)o(hange)g(the)g(parallelism)e(as)i(it)g(mo)o(v)o(es)d(from)i (problem)g(to)h(solution.)27 b(W)l(e)17 b(sho)o(w)i(where)e(v)m(ar-)0 614 y(ious)k(curren)o(t)g(application)f(b)q(enc)o(hmarks)g(apply)h(prob)q(es) h(on)f(the)g(P)o(ath,)h(and)g(whic)o(h)e(agen)o(ts)i(are)0 674 y(b)q(eing)d(measured.)29 b(W)l(e)19 b(recast)g(the)g(traditional)g (time-based)f(p)q(erformance)g(measuremen)o(ts)e(in)o(to)0 734 y(parallelism-based)j(p)q(erformance)g(measuremen)n(ts)f(to)j(sho)o(w)f (that)h(understanding)g(p)q(erformance)0 794 y(implies)11 b(understanding)j (the)g(parallelism.)k(A)13 b(mac)o(hine-indep)q(enden)o(t)e(metho)q(dology)i (is)h(describ)q(ed)0 855 y(for)f(measuring)f(and)h(comparing)f(executed)f (parallelism.)17 b(A)c(sim)o(ulation)d(to)q(ol)k(is)e(used)h(to)g(apply)f (the)0 915 y(metho)q(dology)i(to)h(sev)o(eral)f(of)h(the)f(P)o(erfect)g(Benc) o(hmarks)e(to)j(quan)o(tify)f(their)g(executed)f(parallelism)0 975 y(on)19 b(a)f(single)g(CRA)l(Y)f(Y-MP)h(CPU.)f(Our)h(results)g(suggest)h (that)g(since)e(some)g(of)i(the)e(b)q(enc)o(hmarks)0 1035 y(ha)o(v)o(e)12 b(a)h(similar)d(mix)g(of)j(di\013eren)o(t)f(lev)o(els)e(of)j(parallelism)d (on)j(the)f(Y-MP)l(,)g(the)g(b)q(enc)o(hmarking)f(utilit)o(y)0 1095 y(of)16 b(these)g(programs)h(on)f(the)g(Y-MP)g(ma)o(y)e(b)q(e)j (questioned)e(b)q(ecause)i(these)f(programs)g(exercise)e(the)0 1156 y(mac)o(hine)g(in)i(the)g(same)f(w)o(a)o(y)l(.)73 1216 y(W)l(e)22 b(supp)q(ort)g(the)g(idea)f(that)h(p)q(erformance)f(ev)m(aluation) g(and,)i(in)f(particular,)g(applications-)0 1276 y(lev)o(el)c(b)q(enc)o (hmarking)g(should)i(b)q(ecome)e(more)h(of)h(an)g(analytic,)g(scien)o (ti\014c)e(activit)o(y)l(.)30 b(T)l(o)20 b(do)h(this,)0 1336 y(b)q(etter)h(analytic)h(tec)o(hniques)e(are)h(needed)h(to)g(examine)d(ho)o (w)j(b)q(enc)o(hmarks)f(succeed)f(or)j(fail)e(to)0 1396 y(represen)o(t)13 b(their)h(in)o(tended)f(w)o(orkload)i(and)f(ho)o(w)h(they)f(succeed)f(or)i (fail)f(to)g(test)g(the)g(c)o(haracteristics)0 1457 y(of)i(the)g(mac)o(hines) e(w)o(e)i(use.)21 b(W)l(e)16 b(prop)q(ose)h(that)f(one)h(approac)o(h)f(to)o (w)o(ard)h(these)e(ends)i(is)e(to)i(fo)q(cus)g(on)0 1517 y(parallelism)g(as)j (the)g(essen)o(tial)e(quan)o(tit)o(y)h(and)h(to)g Fi(quantify)g Fm(it.)31 b(P)o(erformance)17 b(ev)m(aluation)j(w)o(ould)0 1577 y(b)q(ecome)15 b(parallelism)f(ev)m(aluation.)73 1637 y(W)l(e)23 b(plan)g(to)g(measure)f(and)h(compare)f(the)h(parallelism)d(c)o (haracteristics)i(of)h(other)g(p)q(opular)0 1697 y(b)q(enc)o(hmark)14 b(sets)i(to)g(quan)o(tify)e(their)h(similarities)d(and)17 b(di\013erences)d (on)i(the)g(CRA)l(Y)e(Y-MP)l(.)h(These)0 1758 y(results)e(can)g(b)q(e)g (merged)f(with)h(the)g(w)o(orkload)g(c)o(haracterization)f(results)h(pro)q (duced)h(b)o(y)e(other)h(stud-)0 1818 y(ies.)24 b(The)18 b(com)o(bined)d (analysis)j(of)f(program)h(and)g(w)o(orkload)g(c)o(haracteristics)e(should)i (help)f(justify)0 1878 y(the)d(inclusion)g(of)h(application)g(b)q(enc)o (hmark)e(programs)i(that)g(are)f(candidates)h(in)f(the)h(construction)0 1938 y(of)k(new)g(b)q(enc)o(hmark)e(sets.)29 b(W)l(e)19 b(w)o(ould)g(also)g (lik)o(e)e(to)i(construct)g(a)h(b)q(enc)o(hmark)d(set)i(based)g(on)g(its)0 1998 y(co)o(v)o(erage)c(of)i(v)m(arious)g(distributions)f(of)g(parallelism,)d (rather)j(than,)h(for)f(example,)e(on)i(application)0 2058 y(areas.)24 b(The)17 b(to)q(ols)h(and)g(metho)q(dology)e(to)i(do)f(this)g (could)g(b)q(e)g(based)h(on)f(the)g(approac)o(h)h(describ)q(ed)0 2119 y(here,)d(and)i(w)o(ould)f(require)f(the)h(understanding)h(of)g(lo)o (w-lev)o(el)d(parallelism)g(on)j(other)f(mac)o(hines.)0 2285 y Fl(7)83 b(Ac)n(kno)n(wledgmen)n(ts)0 2395 y Fm(The)15 b(authors)i(wish)e (to)g(thank)h(the)f(editors)g(and)h(anon)o(ymous)f(referees)f(for)h(their)g (though)o(tful)g(com-)0 2455 y(men)o(ts)g(whic)o(h)g(greatly)h(impro)o(v)o (ed)e(the)i(qualit)o(y)f(of)h(this)g(pap)q(er.)913 2828 y(19)p eop %%Page: 20 26 bop 0 203 a Fl(References)24 313 y Fm([1])24 b(F.)12 b(McMahon,)h(\\The)g (Liv)o(ermore)e(Fortran)i(Kernels:)19 b(A)12 b(test)h(of)g(the)f(n)o (umerical)e(p)q(erformance)100 373 y(range,")17 b(T)l(ec)o(h.)e(Rep.)h (UCRL-53745,)i(La)o(wrence)e(Liv)o(ermore)e(Lab,)j(Dec.)e(1986.)24 475 y([2])24 b(J.)18 b(Dongarra,)i(\\P)o(erformance)d(of)i(v)m(arious)f (computers)f(using)i(standard)g(linear)f(equations,")100 535 y(T)l(ec)o(h.)d(Rep.)h(23,)g(Argonne)h(National)f(Lab,)h(1988.)24 637 y([3])24 b(M.)18 b(Berry)g(et)g(al.,)h(\\The)g(Perfect)f(Club)h(b)q(enc)o (hmarks:)25 b(E\013ectiv)o(e)17 b(p)q(erformance)h(ev)m(aluation)100 697 y(of)23 b(sup)q(ercomputers,")f Fi(International)i(Journal)f(of)g(Sup)n (er)n(c)n(omputer)f(Applic)n(ations)p Fm(,)h(v)o(ol.)e(3,)100 757 y(pp.)16 b(5{40,)i(F)l(all)d(1989.)24 859 y([4])24 b(J.)d(Uniejewski,)f (\\SPEC)i(b)q(enc)o(hmark)d(suite:)30 b(designed)21 b(for)g(to)q(da)o(y's)g (adv)m(anced)g(systems,")100 919 y(T)l(ec)o(h.)15 b(Rep.)h(1,)g(SPEC)h (Newsletter,)d(1989.)24 1021 y([5])24 b(D.)18 b(Bailey)f(et)h(al.,)g(\\The)g (NAS)g(parallel)f(b)q(enc)o(hmarks,")h(T)l(ec)o(h.)f(Rep.)g(RNR-91-002,)j (NASA)100 1081 y(Ames,)14 b(Jan)o(uary)j(1991.)24 1183 y([6])24 b(R.)16 b(P)l(.)f(W)l(eic)o(k)o(er,)e(\\An)j(o)o(v)o(erview)e(of)i(common)e (b)q(enc)o(hmarks,")h Fi(IEEE)i(Computer)p Fm(,)e(pp.)h(65{76,)100 1243 y(1990.)24 1344 y([7])24 b(M.)13 b(Berry)l(,)f(G.)h(Cyb)q(enk)o(o,)h (and)g(J.)f(Larson,)i(\\Scien)o(ti\014c)c(b)q(enc)o(hmark)h(c)o (haracterization,")h Fi(Par-)100 1405 y(al)r(lel)20 b(Computing)p Fm(,)c(v)o(ol.)f(17,)i(pp.)f(1173{1194)q(,)i(Decem)o(b)q(er)c(1991.)24 1506 y([8])24 b(A.)f(v)m(an)h(der)f(Steen,)i(\\Rep)q(ort)f(of)g(the)f(second) h(Eurob)q(en)g(w)o(orkshop)h(on)f(b)q(enc)o(hmarking,")100 1567 y Fi(Sup)n(er)n(c)n(omputer)p Fm(,)15 b(v)o(ol.)h(8,)g(pp.)g(15{19,)h (Septem)o(b)q(er)e(1991.)24 1668 y([9])24 b(J.)e(L.)g(Hennesy)e(and)j(D.)e (A.)g(P)o(atterson,)j Fi(Computer)e(A)o(r)n(chite)n(ctur)n(e:)32 b(A)23 b(Quantitative)h(Ap-)100 1728 y(pr)n(o)n(ach)p Fm(,)15 b(p.)h(36.)22 b(San)16 b(Mateo,)g(CA:)g(Morgan)h(Kaufmann,)e(1990.)0 1830 y([10])24 b(D.)d(J.)f(Kuc)o(k,)h Fi(The)h(Structur)n(e)g(of)f(Computers) h(and)g(Computations)p Fm(,)f(v)o(ol.)f(1,)i(p.)f(100.)36 b(New)100 1890 y(Y)l(ork:)21 b(John)c(Wiley)e(and)i(Sons,)f(1978.)0 1992 y([11])24 b(Cra)o(y)c(Researc)o(h,)g(Inc.,)f Fi(UNICOS)i(Performanc)n(e)g (Utilities)h(R)n(efer)n(enc)n(e)f(Manual)g(SR-2040)g(-)100 2052 y(6.0)p Fm(,)16 b(1991.)0 2154 y([12])24 b(D.)18 b(K.)g(Bradley)g(and)g (J.)g(L.)h(Larson,)h(\\Fine-grained)e(measuremen)o(ts)d(of)k(lo)q(op)g(p)q (erformance)100 2214 y(on)24 b(the)f(CRA)l(Y)f(Y-MP,")g(in)h Fi(Pr)n(o)n(c)n(e)n(e)n(dings)g(of)g(the)h(Fifth)g(SIAM)f(Confer)n(enc)n(e)i (on)f(Par)n(al)r(lel)100 2274 y(Pr)n(o)n(c)n(essing)17 b(for)g(Scienti\014c)j (Computing)p Fm(,)c(Marc)o(h)g(25{27)i(1991.)0 2376 y([13])24 b(D.)19 b(Bradley)l(,)f(G.)g(Cyb)q(enk)o(o,)h(H.)f(Gao,)i(J.)f(Larson,)h(F.)e (Ahmad,)g(J.)h(Golab,)g(and)h(M.)e(Strak)m(a,)100 2436 y(\\Sup)q(ercomputer)d (w)o(orkload)h(decomp)q(osition)f(and)h(analysis,")g(in)g Fi(Pr)n(o)n(c)n(e)n (e)n(dings)f(of)i(the)g(A)o(CM)100 2496 y(International)i(Confer)n(enc)n(e)f (on)g(Sup)n(er)n(c)n(omputing)p Fm(,)e(pp.)g(458{467,)i(June)f(17{21)h(1991.) 0 2598 y([14])24 b(H.)18 b(Gao)h(and)g(J.)f(L.)g(Larson,)i(\\A)e(y)o(ear's)g (pro\014le)g(of)g(sup)q(ercomputer)f(users)i(in)f(di\013eren)o(t)f(ap-)100 2658 y(plication)f(areas)h(using)f(a)h(hardw)o(are)g(p)q(erformance)e (monitor.")g(W)l(ork)h(in)g(progress,)h(1992.)913 2828 y(20)p eop %%Page: 21 27 bop 0 195 a Fm([15])24 b(J.)f(L.)g(Larson,)j(\\Collecting)c(and)i(in)o (terpreting)d(hpm)h(p)q(erformance)g(data)i(on)f(the)g(CRA)l(Y)100 255 y(Y-MP,")16 b Fi(NCSA)i(Datalink)p Fm(,)f(pp.)f(14{24,)i(No)o(v)o(em)o(b) q(er-Dece)o(m)n(b)q(er)13 b(1991.)0 357 y([16])24 b(D.)e(K.)e(Chen,)j(H.)d (M.)h(Su,)i(and)f(P)l(.)f(Y)l(ew,)h(\\The)g(impact)d(of)j(sync)o(hronization) f(and)h(gran)o(u-)100 417 y(larit)o(y)17 b(on)h(parallel)f(systems,")g(in)g Fi(Pr)n(o)n(c)n(e)n(e)n(dings)h(of)h(the)g(17th)g(International)h(Symp)n (osium)e(on)100 477 y(Computer)g(A)o(r)n(chite)n(ctur)n(e)p Fm(,)d(pp.)h(239{249,)i(June)f(1990.)913 2828 y(21)p eop %%Trailer end userdict /end-hook known{end-hook}if %%EOF From owner-pbwg-comm@CS.UTK.EDU Thu Jun 3 11:46:08 1993 Received: from CS.UTK.EDU by netlib2.cs.utk.edu with SMTP (5.61+IDA+UTK-930125/2.8t-UTK) id AA18007; Thu, 3 Jun 93 11:46:08 -0400 Received: from localhost by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10460; Thu, 3 Jun 93 11:46:10 -0400 X-Resent-To: pbwg-comm@CS.UTK.EDU ; Thu, 3 Jun 1993 11:46:06 EDT Errors-To: owner-pbwg-comm@CS.UTK.EDU Received: from sp2.csrd.uiuc.edu by CS.UTK.EDU with SMTP (5.61+IDA+UTK-930125/2.8s-UTK) id AA10426; Thu, 3 Jun 93 11:45:47 -0400 Received: by sp2.csrd.uiuc.edu id AA04359 (5.67a/IDA-1.5); Thu, 3 Jun 1993 10:45:27 -0500 Date: Thu, 3 Jun 1993 10:45:27 -0500 From: "John L. Larson" Message-Id: <199306031545.AA04359@sp2.csrd.uiuc.edu> To: pbwg-comm@cs.utk.edu, perfect.steering@csrd.uiuc.edu Subject: workload paper %!PS-Adobe-2.0 %%Creator: dvips 5.47 Copyright 1986-91 Radical Eye Software %%Title: m1.dvi %%Pages: 25 1 %%BoundingBox: 0 0 612 792 %%EndComments %%BeginProcSet: texc.pro /TeXDict 200 dict def TeXDict begin /N /def load def /B{bind def}N /S /exch load def /X{S N}B /TR /translate load N /isls false N /vsize 10 N /@rigin{ isls{[0 1 -1 0 0 0]concat}if 72 Resolution div 72 VResolution div neg scale Resolution VResolution vsize neg mul TR matrix currentmatrix dup dup 4 get round 4 exch put dup dup 5 get round 5 exch put setmatrix}N /@letter{/vsize 10 N}B /@landscape{/isls true N /vsize -1 N}B /@a4{/vsize 10.6929133858 N}B /@a3{ /vsize 15.5531 N}B /@ledger{/vsize 16 N}B /@legal{/vsize 13 N}B /@manualfeed{ statusdict /manualfeed true put}B /@copies{/#copies X}B /FMat[1 0 0 -1 0 0]N /FBB[0 0 0 0]N /nn 0 N /IE 0 N /ctr 0 N /df-tail{/nn 8 dict N nn begin /FontType 3 N /FontMatrix fntrx N /FontBBox FBB N string /base X array /BitMaps X /BuildChar{CharBuilder}N /Encoding IE N end dup{/foo setfont}2 array copy cvx N load 0 nn put /ctr 0 N[}B /df{/sf 1 N /fntrx FMat N df-tail} B /dfs{div /sf X /fntrx[sf 0 0 sf neg 0 0]N df-tail}B /E{pop nn dup definefont setfont}B /ch-width{ch-data dup length 5 sub get}B /ch-height{ch-data dup length 4 sub get}B /ch-xoff{128 ch-data dup length 3 sub get sub}B /ch-yoff{ ch-data dup length 2 sub get 127 sub}B /ch-dx{ch-data dup length 1 sub get}B /ch-image{ch-data dup type /stringtype ne{ctr get /ctr ctr 1 add N}if}B /id 0 N /rw 0 N /rc 0 N /gp 0 N /cp 0 N /G 0 N /sf 0 N /CharBuilder{save 3 1 roll S dup /base get 2 index get S /BitMaps get S get /ch-data X pop /ctr 0 N ch-dx 0 ch-xoff ch-yoff ch-height sub ch-xoff ch-width add ch-yoff setcachedevice ch-width ch-height true[1 0 0 -1 -.1 ch-xoff sub ch-yoff .1 add]/id ch-image N /rw ch-width 7 add 8 idiv string N /rc 0 N /gp 0 N /cp 0 N{rc 0 ne{rc 1 sub /rc X rw}{G}ifelse}imagemask restore}B /G{{id gp get /gp gp 1 add N dup 18 mod S 18 idiv pl S get exec}loop}B /adv{cp add /cp X}B /chg{rw cp id gp 4 index getinterval putinterval dup gp add /gp X adv}B /nd{/cp 0 N rw exit}B /lsh{rw cp 2 copy get dup 0 eq{pop 1}{dup 255 eq{pop 254}{dup dup add 255 and S 1 and or}ifelse}ifelse put 1 adv}B /rsh{rw cp 2 copy get dup 0 eq{pop 128}{dup 255 eq{pop 127}{dup 2 idiv S 128 and or}ifelse}ifelse put 1 adv}B /clr{rw cp 2 index string putinterval adv}B /set{rw cp fillstr 0 4 index getinterval putinterval adv}B /fillstr 18 string 0 1 17{2 copy 255 put pop}for N /pl[{adv 1 chg}bind{adv 1 chg nd}bind{1 add chg}bind{1 add chg nd}bind{adv lsh}bind{ adv lsh nd}bind{adv rsh}bind{adv rsh nd}bind{1 add adv}bind{/rc X nd}bind{1 add set}bind{1 add clr}bind{adv 2 chg}bind{adv 2 chg nd}bind{pop nd}bind]N /D{ /cc X dup type /stringtype ne{]}if nn /base get cc ctr put nn /BitMaps get S ctr S sf 1 ne{dup dup length 1 sub dup 2 index S get sf div put}if put /ctr ctr 1 add N}B /I{cc 1 add D}B /bop{userdict /bop-hook known{bop-hook}if /SI save N @rigin 0 0 moveto}N /eop{clear SI restore showpage userdict /eop-hook known{eop-hook}if}N /@start{userdict /start-hook known{start-hook}if /VResolution X /Resolution X 1000 div /DVImag X /IE 256 array N 0 1 255{IE S 1 string dup 0 3 index put cvn put}for}N /p /show load N /RMat[1 0 0 -1 0 0]N /BDot 260 string N /rulex 0 N /ruley 0 N /v{/ruley X /rulex X V}B /V statusdict begin /product where{pop product dup length 7 ge{0 7 getinterval (Display)eq}{pop false}ifelse}{false}ifelse end{{gsave TR -.1 -.1 TR 1 1 scale rulex ruley false RMat{BDot}imagemask grestore}}{{gsave TR -.1 -.1 TR rulex ruley scale 1 1 false RMat{BDot}imagemask grestore}}ifelse B /a{moveto}B /delta 0 N /tail{dup /delta X 0 rmoveto}B /M{S p delta add tail}B /b{S p tail} B /c{-4 M}B /d{-3 M}B /e{-2 M}B /f{-1 M}B /g{0 M}B /h{1 M}B /i{2 M}B /j{3 M}B /k{4 M}B /w{0 rmoveto}B /l{p -4 w}B /m{p -3 w}B /n{p -2 w}B /o{p -1 w}B /q{p 1 w}B /r{p 2 w}B /s{p 3 w}B /t{p 4 w}B /x{0 S rmoveto}B /y{3 2 roll p a}B /bos{ /SS save N}B /eos{clear SS restore}B end %%EndProcSet TeXDict begin 1000 300 300 @start /Fa 1 50 df<120C121C12EC120CAFEAFFC00A137D92 11>49 D E /Fb 1 51 df50 D E /Fc 2 64 df<1306131E133E13E6EA03C6EA0F06121C127012E0A21270121C120FEA03C6EA 00E6133E131E13060F127E9113>47 D<7FA5487EA238F9CF80383FFE00EA0FF8EA03E0A2EA0770 EA0630487EEA0808487E11117F9113>63 D E /Fd 3 15 df<1203A4EAE31CEA7338EA1FE0EA07 80A2EA1FE0EA7338EAE31CEA0300A40E107E9013>3 D<7F487EEA0360EA0630487E487E487E48 7E38C00180A238600300EA30066C5A6C5A6C5A6C5A6C5A6C5A11127F9113>5 D14 D E /Fe 4 16 df2 D8 D10 D<121E123FEA7F80EAFFC0A6 EA7F80EA3F00121E0A0C7D8B10>15 D E /Ff 25 119 df<126012F0A212701210A41220A21240 1280040C7C830C>59 D<130113031306A3130CA31318A31330A31360A213C0A3EA0180A3EA0300 A31206A25AA35AA35AA35AA35AA210297E9E15>61 D<140CA2141CA2143C147C145C149C148EEB 010EA213021304A21308A213101320EB3FFEEB4007A21380EA0100A21202A21206121E39FF807F F01C1D7F9C1F>65 D<903801F80890380E0618903838013890386000F048481370485A48C71230 481420120E5A123C15005AA35AA45CA300701302A200305B00385B6C5B6C136038070180D800FE C7FC1D1E7E9C1E>67 D<3801FFE038003C001338A45BA45BA4485AA438038002A31404EA070014 0C14181438000E13F0B5FC171C7E9B1C>76 D79 D<48B5FC39003C03C090383800 E015F01570A24913F0A315E0EBE001EC03C0EC0780EC1E003801FFF001C0C7FCA3485AA448C8FC A45AEAFFE01C1C7E9B1B>I<001FB512F0391C03807039300700300020142012601240130E1280 A2000014005BA45BA45BA45BA41201EA7FFF1C1C7F9B18>84 D<397FF03FE0390F000700000E13 061404A3485BA4485BA4485BA4485BA35CA249C7FCEA60025B6C5AEA1830EA07C01B1D7D9B1C> I97 D<123F1207A2120EA45AA4EA39E0EA3A30EA3C1812381270131CA3EAE038A3133013 70136013C01261EA2300121E0E1D7E9C12>II101 DII105 D<1307130FA213061300A61370139CEA010C1202131C1204 1200A21338A41370A413E0A4EA01C01261EAF180EAF30012E6127C1024809B11>I<39381F81F0 394E20C618394640E81CEB80F0EA8F00008E13E0120EA2391C01C038A315703938038071A215E1 15E23970070064D83003133820127E9124>109 DI<13F8EA03 0CEA0E06487E1218123000701380A238E00700A3130EA25BEA60185BEA30E0EA0F8011127E9114 >I114 DI<13C01201A3EA0380A4EAFFF0EA0700A3120EA45A A4EA3820A21340A2EA1880EA0F000C1A80990F>I<001C13C0EA27011247A238870380A2120EA2 381C0700A438180E20A3EA1C1E380C26403807C38013127E9118>II E /Fg 1 47 df<124012E0124003037D8208>46 D E /Fh 38 122 df 45 D<1238127C12FEA3127C123807077C8610>I<13181378EA01F812FFA21201B3A7387FFFE0A2 13207C9F1C>49 DI<13FE3807FFC0380F07E0381E03F0123FEB81F8A3EA1F0314 F0120014E0EB07C0EB1F803801FE007F380007C0EB01F014F8EB00FCA2003C13FE127EB4FCA314 FCEA7E01007813F8381E07F0380FFFC03801FE0017207E9F1C>I<14E013011303A21307130F13 1FA21337137713E7EA01C71387EA03071207120E120C12181238127012E0B512FEA2380007E0A7 EBFFFEA217207E9F1C>I<1470A214F8A3497EA2497EA3EB06FF80010E7FEB0C3FA201187F141F 01387FEB300FA201607F140701E07F90B5FCA239018001FCA200038090C7FCA20006147FA23AFF E00FFFF8A225227EA12A>65 D67 D70 D72 D I77 D79 DI82 D87 D89 D97 D99 DI<13FE3807FF80380F87C0381E01E0003E13F0EA7C0014F812FC A2B5FCA200FCC7FCA3127CA2127E003E13186C1330380FC0703803FFC0C6130015167E951A>I< EB3F80EBFFC03801E3E0EA03C71207EA0F87EB83C0EB8000A6EAFFFCA2EA0F80B2EA7FF8A21323 7FA211>I<3803FC1E380FFF7F381F0F8F383E07CF383C03C0007C13E0A5003C13C0EA3E07381F 0F80EBFF00EA13FC0030C7FCA21238383FFF806C13F06C13F84813FCEA380048133E00F0131EA4 0078133C007C137C383F01F8380FFFE00001130018217E951C>II<121C123E127FA3123E121CC7FC A7B4FCA2121FB2EAFFE0A20B247EA310>I 107 DI<3AFF07F007F090391FFC1FFC3A1F303E303E 01401340496C487EA201001300AE3BFFE0FFE0FFE0A22B167E9530>I<38FF07E0EB1FF8381F30 7CEB403CEB803EA21300AE39FFE1FFC0A21A167E951F>I<13FE3807FFC0380F83E0381E00F000 3E13F848137CA300FC137EA7007C137CA26C13F8381F01F0380F83E03807FFC03800FE0017167E 951C>I<38FF0FE0EB3FF8381FF07CEB803E497E1580A2EC0FC0A8EC1F80A29038803F00EBC03E EBE0FCEB3FF8EB0FC090C8FCA8EAFFE0A21A207E951F>I114 DI<487EA41203A21207A2120F123FB5FCA2EA0F80ABEB8180A5EB8300EA07 C3EA03FEEA00F811207F9F16>I<38FF01FEA2381F003EAF147E14FE380F81BE3907FF3FC0EA01 FC1A167E951F>I<39FFE01FE0A2390F800600A2EBC00E0007130CEBE01C00031318A26C6C5AA2 6C6C5AA2EB7CC0A2137F6D5AA26DC7FCA2130EA21B167F951E>I<3AFFE7FF07F8A23A1F007800 C0D80F80EB0180147CA23A07C07E030014DE01E05B0003EBDF06EBE18FD801F15B01F3138C9038 FB079C000014D8EBFE03017E13F0A2EB7C01013C5BEB380001185B25167F9528>I<39FFE01FE0 A2390F800600A2EBC00E0007130CEBE01C00031318A26C6C5AA26C6C5AA2EB7CC0A2137F6D5AA2 6DC7FCA2130EA2130CA25B1278EAFC3813305BEA69C0EA7F80001FC8FC1B207F951E>121 D E /Fi 4 50 df0 D3 D15 D49 D E /Fj 31 122 df<130E131E137EEA07FE12FFA212F81200B3AB387FFFFEA317277BA622>49 DII<140E141E143E147E14FEA213011303EB077E130EA2131C1338137013E0A2 EA01C0EA0380EA0700120EA25A5A5A5AB612F8A3C7EAFE00A890387FFFF8A31D277EA622>I<00 0C1303380F803FEBFFFE5C5C5C5C5C49C7FC000EC8FCA6EB7FC0380FFFF8EB80FC380E003E000C 133FC7EA1F8015C0A215E0A21218127C12FEA315C05A0078EB3F80A26CEB7F00381F01FE380FFF F800035BC613801B277DA622>I65 D<91393FF00180903903FFFE03010FEBFF8790393FF007DF9039FF8001FF4848C7127F4848143F D807F0141F000F150F48481407A2485A1603127F5B93C7FC12FFA9127FA26DEC0380123FA26C7E EE07006C7E0007150ED803FC141E6C6C5C6C6C6C13F890393FF007E0010FB55A010391C7FC9038 003FF829297CA832>67 D70 D73 D77 D82 D87 D<48B47E000713F0380F81F8381FC07EA280D80F801380EA0700C7FCA3EB0FFF90B5FC38 07FC3FEA0FE0EA3F8013005A12FEA4007E137F007F13DF393F839FFC380FFF0F3801FC031E1B7E 9A21>97 D 99 DI II<9038FF81F00003EBE7F8390FC1FE7C381F80FC9038007C3848EB7E1048EB7F 00A66C137E6C137CEB80FC380FC1F8381FFFE0001813800038C8FCA2123C123E383FFFF86C13FF 15806C14C06C14E0001F14F0383E000748EB01F8481300A4007CEB01F0003C14E0001FEB07C039 0FC01F803903FFFE0038007FF01E287E9A22>II<1207EA0F80EA1FC0EA3FE0 A3EA1FC0EA0F80EA0700C7FCA7EAFFE0A3120FB3A3EAFFFEA30F2B7DAA14>I107 DI<3BFFC07F800FF0903AC1FFE03FFC903AC383F0707E3B0FC603F8C07F903ACC01 F9803F01D8D9FF00138001F05BA201E05BB03CFFFE1FFFC3FFF8A3351B7D9A3A>I<38FFC07F90 38C1FFC09038C787E0390FCE03F013D88113F0A213E0B03AFFFE3FFF80A3211B7D9A26>II<38FFC1F0EB C7FCEBCE3E380FD87FA213F0143E141CEBE000B0B5FCA3181B7E9A1C>114 D<3803FE30380FFFF0EA1E03EA380048137012F0A27E6C1300EAFFE0EA7FFEEBFF806C13E06C13 F0000713F8C6FCEB03FC13000060137C00E0133C7E14387E6C137038FF01E038F7FFC000C11300 161B7E9A1B>I<1370A413F0A312011203A21207381FFFF0B5FCA23807F000AD1438A61203EBF8 70000113603800FFC0EB1F8015267FA51B>I<39FFE03FF8A3000F1303B214071207140F3A03F0 3BFF803801FFF338003FC3211B7D9A26>I<3BFFFE7FFC0FFEA33B0FE007E000E03B07F003F001 C0A29039F807F80300031680A23B01FC0EFC0700A2D9FE1E5B000090381C7E0EA29039FF383F1E 017F141C0278133C90393FF01FB8A216F86D486C5AA26D486C5AA36D486C5AA22F1B7F9A32> 119 D<3AFFFE03FF80A33A07F0007000A26D13F000035CEBFC0100015CA26C6C485AA2D97F07C7 FCA2148FEB3F8E14DEEB1FDCA2EB0FF8A36D5AA26D5AA26D5AA2495AA2EA3807007C90C8FCEAFE 0F130E131E5BEA7C78EA3FE0EA0FC021277F9A24>121 D E /Fk 44 122 df12 D<1218123CA31204A21208A21210122012401280060C779C0D>39 D<12181238127812381208A2 1210A212201240A21280050C7D830D>44 D<1230127812F0126005047C830D>46 D<133C13C6EA0183EA030312061480120E120C121C1400485AA4EA700EA4485AA35BA21330485A 12E0EA60C0EA7180001EC7FC111D7B9B15>48 D<133C13C6EA0183EA03031206120E000C138012 1C1400A2485AA35B130EEA181EA2EA0C6CEA079CEA001C1318133813305BEAE0C0A2EA81800086 C7FC127C111D7B9B15>57 D<1418A21438A21478A214B8130114381302143CEB041CA213081318 13101320A2EB7FFCEB401C1380120113001202A2487F120C001C131EB4EBFFC01A1D7E9C1F>65 D<903803F02090381E0C6090383002E09038E003C03801C001EA038048C7FC000E1480121E121C 123C15005AA35AA41404A35C12705C6C5B00185B6C485AD80706C7FCEA01F81B1E7A9C1E>67 D<3801FFFE39003C078090383801C0A2EC00E0A24913F01570A215F05BA43901C001E0A315C038 0380031580140715003807000E5C5C5C380E01C0B5C7FC1C1C7D9B1F>I<48B512E038003C0001 3813601540A35BA214201500495AA214C013FF3801C080A43803810113801402A248485AA2140C 5C000E1378B55A1B1C7D9B1C>I<48B512C038003C01EB38001580A35BA214201500495AA214C0 13FF3801C080A4D80381C7FC1380A348C8FCA45AEAFFF01A1C7D9B1B>I<3801FFC038003C0013 38A45BA45BA4485AA4485AA448C7FCA45AEAFFE0121C7E9B10>73 DI<3801FFE038003C 001338A45BA45BA4485AA438038008A31410EA07001430146014E0380E03C0B5FC151C7D9B1A> 76 DI<3901FC03FE39001C0070013C1360012E1340A301471380A3EB43809038838100 A2138114C1380101C2A2EB00E2A2000213E41474A3481338A3000C1318001C1310EAFF801F1C7D 9B1F>I<3801FFFC38003C079038380380EC01C0A3EB7003A31580EBE0071500140E14383801FF E001C0C7FCA3485AA448C8FCA45AEAFFE01A1C7D9B1C>80 D<3801FFF838003C0EEB3807EC0380 A3EB7007A3EC0F00EBE00E5C1470EBFFC0EA01C014601470A2EA0380A4380700F01540A2158048 137839FFE07900C7121E1A1D7D9B1E>82 DI<001FB512C0381C070138300E0000201480126012405B 1280A2000014005BA45BA45BA4485AA41203EA7FFE1A1C799B1E>I<397FF0FF80390F001C0000 0E13181410A3485BA4485BA4485BA44848C7FCA31302A25BA2EA6008EA3030EA1040EA0F80191D 779B1F>I97 D<123F1207A2120EA45AA4EA39C0EA3E60EA3830A2EA7038A4EAE070A3 136013E0EAC0C012C1EA6180EA6300123C0D1D7B9C13>IIIII<13F3EA018FEA030FEA0607EA0E0E12 0C121CA2EA381CA413381230A2EA187813F0EA0F701200A213E0A2EAC0C012E1EAC300127E101A 7D9113>III107 DI<393C1E078039266318C0394683A0E0384703C0008E1380A2120EA2391C0701C0A3EC03 80D8380E1388A2EC0708151039701C032039300C01C01D127C9122>IIII114 DI<13C01201A3 EA0380A4EAFFE0EA0700A3120EA45AA4EA3840A31380EA1900120E0B1A7D990E>III<381E0183382703871247148338870701A2120EA2381C0E02A31404EA180C131C140800 1C1310380C26303807C3C018127C911C>III E /Fl 85 124 df11 D<137E3801C180EA0301380703C0120E EB018090C7FCA5B512C0EA0E01B0387F87F8151D809C17>II<90383F07E03901C09C18380380F0D80701133C000E 13E00100131892C7FCA5B612FC390E00E01CB03A7FC7FCFF80211D809C23>I34 D<000F14C0EA18803930 6003803970380700386027FB38E010065CA25CA25C5CEA602000705B38304180EA1881390F0303 C03900060620EC0C1090380C1C08EB1818EC380413301360A213C0A2EA01803903001808141C00 06EB0C1048EB06200004EB03C01E217E9E23>37 D<13E0EA0190EA0308A21207A45BA25B5B3903 C03FE09038800F0014061404EA05C000095B3810E010123038607020A238E03840133CEB1C8090 380F00207F0070EB8040383009C0391830E1803907C03E001B1F7E9D20>I<126012F012F81268 1208A31210A2122012401280050C7C9C0C>I<13401380EA0100120212065AA25AA25AA2127012 60A312E0AC1260A312701230A27EA27EA27E12027EEA008013400A2A7D9E10>I<7E12407E7E12 187EA27EA27EA213801201A313C0AC1380A312031300A21206A25AA25A12105A5A5A0A2A7E9E10 >II<1306ADB612E0A2D80006C7FCAD1B1C7E9720>I<126012F0A212701210A41220A212 401280040C7C830C>II<126012F0A2126004047C830C>I<1301130313 06A3130CA31318A31330A31360A213C0A3EA0180A3EA0300A31206A25AA35AA35AA35AA35AA210 297E9E15>II<12035A123F12C71207B3A4EA0F80EAFFF80D1C7C9B 15>III<130CA2131C133CA2135C13DC13 9CEA011C120312021204120C1208121012301220124012C0B512C038001C00A73801FFC0121C7F 9B15>II<13F0EA030CEA0604EA0C0E EA181E1230130CEA7000A21260EAE3E0EAE430EAE818EAF00C130EEAE0061307A51260A2EA7006 EA300E130CEA1818EA0C30EA03E0101D7E9B15>I<1240387FFF801400A2EA4002485AA25B485A A25B1360134013C0A212015BA21203A41207A66CC7FC111D7E9B15>III<126012F0A212601200AA126012F0A2126004127C 910C>I<007FB512C0B612E0C9FCA8B612E06C14C01B0C7E8F20>61 D<1306A3130FA3EB1780A3 EB23C0A3EB41E0A3EB80F0A200017FEB0078EBFFF83803007C0002133CA20006133E0004131EA2 000C131F121E39FF80FFF01C1D7F9C1F>65 DI<9038 1F8080EBE0613801801938070007000E13035A14015A00781300A2127000F01400A80070148012 78A212386CEB0100A26C13026C5B380180083800E030EB1FC0191E7E9C1E>IIII<90381F8080EBE06138018019 38070007000E13035A14015A00781300A2127000F01400A6ECFFF0EC0F80007013071278A21238 7EA27E6C130B380180113800E06090381F80001C1E7E9C21>I<39FFF3FFC0390F003C00ACEBFF FCEB003CAD39FFF3FFC01A1C7E9B1F>III<39FFF03FE0390F000F0014 0C14085C5C5C5C49C7FC13025B130E131F132FEB27801347EB83C0EB01E0A26D7E80147880A280 141F158039FFF07FF01C1C7E9B20>IIIIIIII<3807E080EA1C19EA3005EA7003EA600112E01300A36C13007E127CEA7FC0EA 3FF8EA1FFEEA07FFC61380130FEB07C0130313011280A300C01380A238E00300EAD002EACC0CEA 83F8121E7E9C17>I<007FB512C038700F010060130000401440A200C014201280A300001400B1 497E3803FFFC1B1C7F9B1E>I<39FFF07FC0390F000E001404B3A26C5B138000035B12016C6C5A EB70C0011FC7FC1A1D7E9B1F>I<39FFE00FF0391F0003C06CEB018015006D5A00071302A26C6C 5AA36C6C5AA213F000005BA2EBF830EB7820A26D5AA36D5AA2131F6DC7FCA21306A31C1D7F9B1F >I<3AFFE0FFE0FF3A1F001F003C001E011E13186C011F1310A3D807801420EC2780A2D803C014 40EC43C0A213E00001903881E080A33A00F100F100A3017913FA017A137AA2013E137C013C133C A301181318A3281D7F9B2B>I<397FF0FFC0390FC03E0038078018EA03C0EBE01000015BEBF060 00001340EB7880137D013DC7FC7F131F7F80A2EB13C0EB23E01321EB41F0EBC0F8EB8078380100 7C48133C00027F0006131F001FEB3F8039FFC0FFF01C1C7F9B1F>I<39FFF007FC390F8001E000 07EB0080EBC00100031400EBE002EA01F000005B13F8EB7808EB7C18EB3C106D5A131F6D5A14C0 6D5AABEB7FF81E1C809B1F>I<387FFFF0EA7C01007013E0386003C0A238400780130F1400131E 12005B137C13785BA2485A1203EBC010EA0780A2EA0F00481330001E13205A14604813E0EAF803 B5FC141C7E9B19>I<12FEA212C0B3B312FEA207297C9E0C>II<12FEA21206B3B312FEA20729809E0C>I<12 08121012201240A21280A312B012F812781230050C7D9C0C>96 DI<12 FC121CAA137CEA1D86EA1E03381C018014C0130014E0A614C013011480381E0300EA1906EA10F8 131D7F9C17>II<133F1307AAEA03E7EA0C17EA180F487E1270126012E0A61260127012 306C5AEA0C373807C7E0131D7E9C17>II<13F8EA018CEA071E1206EA 0E0C1300A6EAFFE0EA0E00B0EA7FE00F1D809C0D>II<12FC121CAA137C1387EA1D03001E13 80121CAD38FF9FF0141D7F9C17>I<1218123CA21218C7FCA712FC121CB0EAFF80091D7F9C0C>I< 13C0EA01E0A2EA00C01300A7EA0FE01200B3A21260EAF0C012F1EA6180EA3E000B25839C0D>I< 12FC121CAAEB3FC0EB0F00130C13085B5B5B13E0121DEA1E70EA1C781338133C131C7F130F1480 38FF9FE0131D7F9C16>I<12FC121CB3A9EAFF80091D7F9C0C>I<39FC7E07E0391C838838391D01 9018001EEBE01C001C13C0AD3AFF8FF8FF8021127F9124>IIIIIII<1204A4120CA2121C123CEAFFE0EA1C00A91310A5120CEA0E20EA03C00C1A7F9910>I< 38FC1F80EA1C03AD1307120CEA0E1B3803E3F014127F9117>I<38FF07E0383C0380381C0100A2 EA0E02A26C5AA3EA0388A213D8EA01D0A2EA00E0A3134013127F9116>I<39FF3FCFE0393C0F03 80381C07011500130B000E1382A21311000713C4A213203803A0E8A2EBC06800011370A2EB8030 000013201B127F911E>I<387F8FF0380F03801400EA0702EA0384EA01C813D8EA00F013701378 13F8139CEA010E1202EA060738040380381E07C038FF0FF81512809116>I<38FF07E0383C0380 381C0100A2EA0E02A26C5AA3EA0388A213D8EA01D0A2EA00E0A31340A25BA212F000F1C7FC12F3 1266123C131A7F9116>III E /Fm 54 122 df<90383FE3F83901F03F1C3903C03E3E0007137CEA0F80151C1500A5B612C0A2 390F807C00AE397FE1FFC0A21F1D809C1C>11 D13 D37 D<13201340EA0180120313001206120E5AA2123C1238A21278A312F85AA97E1278A31238A2123C 121CA27E12067E13801201EA004013200B297C9E13>40 D<7E12401230123812187E120E7EA213 801203A213C0A313E01201A9120313C0A31380A212071300A2120E120C5A1238123012405A0B29 7D9E13>I43 D<127812FCA212FEA2127A1202 A21204A21208A212301240070E7D850D>II<127812FCA4127806067D85 0D>III<1360EA01E0120F12FF 12F31203B3A2387FFF80A2111B7D9A18>IIII<38180180381FFF005B5B5B13C0 0018C7FCA4EA19F8EA1E0E38180F80EA1007000013C014E0A3127812F8A214C012F038600F8038 381F00EA1FFEEA07F0131B7E9A18>I<137EEA03FF38078180380F03C0EA1E07123CEB038048C7 FCA212F813F8EAFB0E38FA0780EAFC0314C000F813E0A41278A214C0123CEB0780381E0F00EA07 FEEA03F8131B7E9A18>I<1260387FFFE0A214C01480A238E00300EAC0065B5BC65AA25B13E0A2 12015B1203A41207A66C5A131C7D9B18>III65 DI<90381FE02090 38FFF8E03803F80F3807C003380F800148C7FC123E1560127E127C00FC1400A8007C1460127E12 3E15C07E390F8001803907C003003803F80E3800FFFCEB1FE01B1C7D9B22>II70 D73 D77 D79 DI82 D<3807F820381FFEE0EA3C07EA7801EA700012F01460A26C130012FEEAFFE0EA7FFE6C7E148000 0F13C06C13E0EA007FEB03F01301130012C0A214E07E38F001C0EAFC0338EFFF00EA83FC141C7D 9B1B>I<39FFFC03FFA2390FC00030B3120715606C6C13E03901F001C03900FC078090387FFE00 EB0FF8201C7E9B25>85 D97 DIIII<137F3801E3803803C7C0EA0787120FEB8380EB 8000A5EAFFF8A2EA0F80AEEA7FF8A2121D809C0F>I<3803F8F0380E0F38121E381C0730003C13 80A4001C1300EA1E0FEA0E0EEA1BF80010C7FC1218A2EA1FFF14C06C13E04813F0387801F838F0 0078A300701370007813F0381E03C03807FF00151B7F9118>II<121E123FA4121EC7FCA6B4FCA2121FAEEAFF E0A20B1E7F9D0E>I<137813FCA413781300A6EA03FCA2EA007CB2127012F8137813F0EA70E0EA 1F800E26839D0F>III<39FF 0FC07E903831E18F3A1F40F20780D980FC13C0A2EB00F8AB3AFFE7FF3FF8A225127F9128>I<38 FF0FC0EB31E0381F40F0EB80F8A21300AB38FFE7FFA218127F911B>II<38FF3F80EBE1E0381F80F0EB0078147C143C143EA6143C147C1478EB80F0EBC1E0EB3F 0090C7FCA6EAFFE0A2171A7F911B>I114 DI<1203A45AA25AA2EA3FFC12FFEA1F00A913 0CA4EA0F08EA0798EA03F00E1A7F9913>I<38FF07F8A2EA1F00AC1301120F380786FFEA01F818 127F911B>I<38FFC1FCA2381F0060EB80E0000F13C013C03807C180A23803E300A2EA01F6A213 FE6C5AA21378A2133016127F9119>I<38FFC1FCA2381F0060EB80E0000F13C013C03807C180A2 3803E300A2EA01F713F6EA00FE5BA21378A21330A21370EA706012F85BEAF9800073C7FC123E16 1A7F9119>121 D E /Fn 60 121 df<13FFEA0387EA07071206120EA6B5FCEA0E07AE387F9FE0 131A809915>13 D38 D<1380EA010012025A120C120812185AA35AA412E0AA1260A47EA37E1208120C 12047E7EEA008009267D9B0F>40 D<7E12407E7E12181208120C7EA37EA41380AA1300A41206A3 5A1208121812105A5A5A09267E9B0F>I<126012F0A212701210A31220A212401280040B7D830B> 44 DI<126012F0A2126004047D830B>I48 D<12035AB4FC1207B3A2 EAFFF00C187D9713>III<1330A2137013F0A2EA017012031202120412 0C1208121012301220124012C0B5FCEA0070A6EA07FF10187F9713>III<1240EA7FFE13FCA2EA 4008EA8010A21320EA0040A213801201A213005AA45AA612020F197E9813>III<126012F0A212601200A8126012F0A2126004107D8F0B>I<137F380180C03806003000081308 487F38203E0213E13841C081384380710083EB7080EA8700A6EA838012433941C0F1003820E131 EB3E1E6CC8FC7E0006EB03803901803E0038007FE0191A7E991E>64 D<130CA3131EA2133F1327 A2EB4380A3EB81C0A348C67EA213FF38020070A20006137800041338A2487FA2001C131EB4EBFF C01A1A7F991D>IIIIII<38FFE7FF380E0070AB380FFFF0380E0070AC38FFE7FF181A7E991D>72 DI<39FFE07F80390E001E00141814105C5C5C49C7FC 13025B5B131C132E134E1387380F0380120E6D7E6D7EA21470A28080143E39FFE0FF80191A7E99 1E>75 DII<38FE01FF380F00381410EA0B80A2EA09C0EA08E0A21370A21338131C A2130EA21307EB0390A2EB01D0A2EB00F01470A21430121C38FF8010181A7E991D>I<137F3801 C1C038070070000E7F487F003C131E0038130E0078130F00707F00F01480A80078EB0F00A20038 130E003C131E001C131C6C5B6C5B3801C1C0D8007FC7FC191A7E991E>II82 DI<007F B5FC38701C0700401301A200C0148000801300A300001400B13803FFE0191A7F991C>I<38FFE1 FF380E00381410B20006132012076C1340EA01803800C180EB3E00181A7E991D>I<39FF801FE0 391E000700000E1306000F13046C5B13806C6C5A00011330EBE0206C6C5A1370EB78801338011D C7FC131F130EAAEBFFE01B1A7F991D>89 D97 D<12FC121CA913F8EA1F0EEA1E07381C 0380130114C0A6EB03801400EA1E07EA1B0CEA10F0121A7F9915>II<137E130EA9EA03CEEA0C3EEA38 0E1230127012E0A612601270EA381EEA1C2E3807CFC0121A7F9915>IIII<12FC121CA913F8EA1D0CEA1E0EA2121CAB38FF9FC0121A7F9915>I<121812 3CA21218C7FCA612FC121CAEEAFF80091A80990A>I<12FC121CA9EB3F80EB1E00131813105B5B EA1DC0EA1FE0121C1370137813387F131E131F38FF3FC0121A7F9914>107 D<12FC121CB3A6EAFF80091A80990A>I<38FC7C1F391D8E6380391E0781C0A2001C1301AB39FF 9FE7F81D107F8F20>IIII114 DI<1204A3 120CA2121C123CEAFFC0EA1C00A81320A5EA0E40EA03800B177F960F>II<38FF1F80383C0600EA1C04A2EA1E0CEA0E08A26C5AA21390EA 03A0A2EA01C0A36C5A11107F8F14>I<39FF3F9F80393C0E070000381306381C16041317001E13 0C380E23081488000F13983807419014D03803C1E01380A200015BEB004019107F8F1C>I<38FF 3F80383C1C00EA1C18EA0E106C5A13606C5A12017F1203EA0270487E1208EA181CEA381E38FC3F C012107F8F14>I E /Fo 2 51 df<120C123C12CC120CACEAFF8009107E8F0F>49 D<121FEA6180EA40C0EA806012C01200A213C0EA0180EA030012065AEA10201220EA7FC012FF0B 107F8F0F>I E /Fp 30 122 df<126012F0A212701210A21220A21240A2040A7D830A>44 DI<126012F0A2126004047D830A>I<12035AB4FC1207B1EA7FF00C157E 9412>49 DI<13101338A3135CA3138EA3EA0107A2 00031380EA0203A23807FFC0EA0401A2380800E0A21218003813F038FE03FE17177F961A>65 D67 D<38FF83FE381C0070AA381FFFF038 1C0070AA38FF83FE17177F961A>72 D97 D<12FC121CA813F8EA1F06EA1C031480130114C0 A4148013031400EA1B0EEA10F81217809614>II<137E130EA8EA07CEEA1C3EEA300E1270126012E0A4 12601270EA301EEA182E3807CFC012177F9614>IIII<12FC121C A8137CEA1D8EEA1E07121CAA38FF9FE01317809614>I<1218123CA212181200A5127C121CAC12 FF081780960A>I<12FC121CB3A3EAFF80091780960A>108 D<38FC7C1F391D8E6380391E0781C0 001C1301AA39FF9FE7F81D0E808D1E>II< EA07C0EA1830EA3018EA600CA2EAE00EA5EA701CEA3018EA1830EA07C00F0E7F8D12>II< EA07C2EA1C26EA381EEA700E126012E0A412601270EA301EEA1C2EEA07CEEA000EA5EB7FC01214 7F8D13>III<1208A31218A212 38EAFF80EA3800A71340A4EA1C80EA0F000A147F930E>III<38FCFE7C383838381410381C3C20A2134C380E4E40A2138638078780A2130300031300A2 160E7F8D19>I121 D E /Fq 54 121 df<132013401380EA01005A12061204120C A25AA25AA312701260A312E0AE1260A312701230A37EA27EA2120412067E7EEA0080134013200B 327CA413>40 D<7E12407E7E12187E12041206A27EA2EA0180A313C01200A313E0AE13C0A31201 1380A3EA0300A21206A21204120C5A12105A5A5A0B327DA413>I<127012F812FCA212741204A4 1208A21210A212201240060F7C840E>44 DI<127012F8A3127005057C 840E>I48 D<13801203120F12F31203B3A9EA07 C0EAFFFE0F217CA018>III<13021306130EA2131EA2132E13 4EA2138EA2EA010E1202A21204A212081210A21220A212401280B512F838000E00A7131F3801FF F015217FA018>I<00101380381E0700EA1FFF5B13F8EA13E00010C7FCA613F8EA130EEA140738 1803801210380001C0A214E0A4127012F0A200E013C01280EA4003148038200700EA1006EA0C1C EA03F013227EA018>I<137EEA01C138030080380601C0EA0E03121C381801800038C7FCA21278 1270A2EAF0F8EAF30CEAF4067F00F81380EB01C012F014E0A51270A3003813C0A238180380001C 1300EA0C06EA070CEA01F013227EA018>I<12401260387FFFE014C0A23840008038C001001280 1302A2485A5BA25B133013201360A313E05BA21201A41203A86C5A13237DA118>III<127012F8A312701200AB127012F8A3127005157C940E>I64 D<497EA3497EA3EB05E0A2EB0DF01308A2497E1478A2497EA3497EA3497EA290B5FC3901000780 A24814C000021303A24814E01401A2000CEB00F0A2003EEB01F839FF800FFF20237EA225>I<90 3807E0109038381830EBE0063901C0017039038000F048C7FC000E1470121E001C1430123CA200 7C14101278A200F81400A812781510127C123CA2001C1420121E000E14407E6C6C13803901C001 003800E002EB381CEB07E01C247DA223>67 DIII<903807F00890383C0C18EBE0023901C001B839038000F848C71278 481438121E15185AA2007C14081278A200F81400A7EC1FFF0078EB00F81578127C123CA27EA27E 7E6C6C13B86C7E3900E0031890383C0C08903807F00020247DA226>I<39FFFC3FFF390FC003F0 39078001E0AE90B5FCEB8001AF390FC003F039FFFC3FFF20227EA125>II<3803FFF038001F007FB3A6127012F8A2130EEAF01EEA401C 6C5AEA1870EA07C014237EA119>I76 DI82 D<3803F020380C0C60EA1802383001E0EA70000060136012E0A21420 A36C1300A21278127FEA3FF0EA1FFE6C7E0003138038003FC0EB07E01301EB00F0A214707EA46C 1360A26C13C07E38C8018038C60700EA81FC14247DA21B>I<007FB512F8397807807800601418 00401408A300C0140C00801404A400001400B3A3497E0003B5FC1E227EA123>I<39FFFC07FF39 0FC000F86C4813701520B3A5000314407FA2000114806C7E9038600100EB3006EB1C08EB03F020 237EA125>I<3BFFF03FFC03FE3B1F8007E000F86C486C4813701720A26C6C6C6C1340A32703C0 02F01380A33B01E004780100A33A00F0083C02A39039F8183E06903978101E04A2137C90393C20 0F08A390391E400790A390390F8003E0A36D486C5AA36D5C010213002F237FA132>87 D97 D<120E12FE121E120EAB131FEB61C0EB8060380F00 30000E1338143C141C141EA7141C143C1438000F1370380C8060EB41C038083F0017237FA21B> II<14E0130F13011300ABEA01F8EA0704EA0C02EA1C01EA3800 1278127012F0A7127012781238EA1801EA0C0238070CF03801F0FE17237EA21B>II<133C13C6EA018F1203130FEA0700A9EAFFF8EA0700B213 80EA7FF8102380A20F>I<14703801F19838071E18EA0E0E381C0700A2003C1380A4001C1300A2 EA0E0EEA0F1CEA19F00010C7FCA21218A2EA1FFE380FFFC014E0383800F0006013300040131812 C0A300601330A2003813E0380E03803803FE0015217F9518>I<120E12FE121E120EABEB1F80EB 60C0EB80E0380F0070A2120EAF38FFE7FF18237FA21B>I<121C123EA3121CC7FCA8120E12FE12 1E120EB1EAFFC00A227FA10E>II<120E12FE121E120EB3ADEAFFE00B237FA20E>108 D<390E1FC07F3AFE60E183803A1E807201C03A0F003C00E0A2000E1338AF3AFFE3FF8FFE27157F 942A>I<380E1F8038FE60C0381E80E0380F0070A2120EAF38FFE7FF18157F941B>III114 DI<1202A41206A3120E121E123EEAFFF8EA0E 00AB1304A6EA07081203EA01F00E1F7F9E13>I<000E137038FE07F0EA1E00000E1370AD14F0A2 38060170380382783800FC7F18157F941B>I<38FFC1FE381E0078000E13301420A26C1340A238 038080A33801C100A2EA00E2A31374A21338A3131017157F941A>I<38FF83FE381F01F0380E00 C06C1380380381001383EA01C2EA00E41378A21338133C134E138EEA0187EB0380380201C00004 13E0EA0C00383E01F038FF03FE17157F941A>120 D E /Fr 29 120 df12 D<127812FCA212FEA2127A1202A51204A31208A212101220 A2124007147AB112>39 D<1403A34A7EA24A7EA3EC17E01413A2EC23F01421A2EC40F8A3EC807C A20101137EEC003EA20102133F81A2496D7EA3496D7EA2011880011FB5FCA29039200003F01501 A249801500A249147CA348C87EA248153F825AD81F80EC3F80D8FFE0903803FFFCA22E327EB132 >65 D<91383FE001903901FFF803903807F01E90391F800307013EC712870178144F49142F4848 141F4848140F485A000F150790C8FC481503121E123E003C1501127CA30078150012F8AB127812 7C1601A2123C123E121E001F15027E6D1406000715046C6C14086C7E6C6C141001781420013E14 C090391F800380903907F00F00903801FFFC9038003FE028337CB130>67 D72 D77 D80 D82 D<90387F80203801FFE03907C07860380F001C 001EEB06E048130300381301007813001270156012F0A21520A37E1500127C127E7E13C0EA1FF8 6CB47E6C13F86C7FC613FF010F1380010013C0EC1FE01407EC03F01401140015F8A26C1478A57E 15706C14F015E07E6CEB01C000ECEB038000C7EB070038C1F01E38807FFCEB0FF01D337CB125> I85 D89 D<13FE380303C0380C00E00010 137080003C133C003E131C141EA21208C7FCA3EB0FFEEBFC1EEA03E0EA0F80EA1F00123E123C12 7C481404A3143EA21278007C135E6CEB8F08390F0307F03903FC03E01E1F7D9E21>97 D99 DIII<15F090387F03083901C1C41C380380E8390700700848EB 7800001E7FA2003E133EA6001E133CA26C5B6C13706D5A3809C1C0D8087FC7FC0018C8FCA5121C 7E380FFFF86C13FF6C1480390E000FC00018EB01E048EB00F000701470481438A500701470A26C 14E06CEB01C00007EB07003801C01C38003FE01E2F7E9F21>II<120FEA1F80A4EA0F00C7FCABEA078012FFA2120F1207B3A6EA0FC0EAFFF8A20D307EAF 12>I108 D<260780FEEB1FC03BFF83078060F0903A8C03C180783B0F9001E2003CD807A013E4DA00F47F01 C013F8A2495BB3A2486C486C133F3CFFFC1FFF83FFF0A2341F7E9E38>I<380780FE39FF830780 90388C03C0390F9001E0EA07A06E7E13C0A25BB3A2486C487E3AFFFC1FFF80A2211F7E9E25>I< EB1FC0EBF0783801C01C38070007481480001EEB03C0001C1301003C14E0A248EB00F0A300F814 F8A8007814F0007C1301003C14E0A26CEB03C0A26CEB07803907800F003801C01C3800F078EB1F C01D1F7E9E21>I<380781FC38FF860790388803C0390F9001E03907A000F001C013785B153CA2 153E151E151FA9153EA2153C157C15786D13F013A0EC01E090389803809038860F00EB81F80180 C7FCAB487EEAFFFCA2202D7E9E25>I<380783E038FF8C18EB907C120FEA07A0EBC0381400A35B B3487EEAFFFEA2161F7E9E19>114 D<3801FC10380E0330381800F048137048133012E01410A3 7E6C1300127EEA3FF06CB4FC6C13C0000313E038003FF0EB01F813006C133CA2141C7EA27E1418 6C1338143000CC136038C301C03880FE00161F7E9E1A>I<1340A513C0A31201A212031207120F 381FFFE0B5FC3803C000B01410A80001132013E000001340EB78C0EB1F00142C7FAB19>II<3BFFF07FF80FFCA23B0FC007C003F0D98003EB01C00007ED0080A2D803C090 38E00100A214073A01E004F002A2EC0870D800F0EB7804A2EC10380178EB3C08A2EC201C013CEB 1E10A2EC400E011EEB0F20A2EC8007010F14C0A2EC00036D5CA201061301010291C7FC2E1F7F9E 30>119 D E end %%EndProlog %%BeginSetup %%Feature: *Resolution 300 TeXDict begin %%EndSetup %%Page: 0 1 bop 227 893 a Fr(A)21 b(Y)-6 b(ear's)22 b(Pro\014le)g(of)f(Academic)g(Sup)r (ercomputer)g(Users)249 984 y(Using)f(the)i(CRA)-6 b(Y)22 b(Hardw)n(are)f(P)n (erformance)h(Monitor)462 1167 y Fq(Hui)15 b(Gao)658 1148 y Fp(1)279 1225 y Fq(Cen)o(ter)h(for)g(Sup)q(ercomputing)281 1283 y(Researc)o(h)g(and)g(Dev)o(elopmen)o(t)256 1341 y(465)i(CSRL,)e(1308)i (W.)e(Main)g(St.)382 1399 y(Urbana,)h(IL)f(61801)281 1457 y(Email:)j (hgao@csrd.uiuc.edu)362 1515 y(T)l(el:)h(\(217\)-244-5)q(287)356 1573 y(F)l(ax:)h(\(217\)-244-13)q(51)1170 1167 y(John)c(L.)f(Larson)1511 1148 y Fp(2)1060 1225 y Fq(Cen)o(ter)f(for)i(Sup)q(ercomputing)1062 1283 y(Researc)o(h)e(and)i(Dev)o(elopmen)o(t)1037 1341 y(465)g(CSRL,)g(1308)g (W.)f(Main)g(St.)1163 1399 y(Urbana,)g(IL)h(61801)1040 1457 y(Email:)j(jlarson@csrd.uiuc.edu)1142 1515 y(T)l(el:)h(\(217\)-244-590)q(8) 1136 1573 y(F)l(ax:)h(\(217\)-244-135)q(1)821 1692 y(April)15 b(8,)h(1993)-38 2596 y Fo(1)-21 2612 y Fn(Curren)o(t)e(address:)k(Kuc)o(k)13 b(&)g(Asso)q(ciates,)i(Inc.,)d(1906)i(F)m(o)o(x)f(Driv)o(e,)h(Champaign,)h (IL)e(61820,)h(Email:)19 b(hgao@k)n(ai.com.)g(T)m(el:)e(\(217\)-356-)-90 2658 y(2288)d(,)e(F)m(ax:)17 b(\(217\)-356-5199)-38 2688 y Fo(2)-21 2704 y Fn(presen)o(ting)e(author)p eop %%Page: 0 2 bop 869 713 a Fm(Abstract)-90 854 y Fl(This)13 b(pap)q(er)h(describ)q(es)h (some)d(preliminary)f(results)j(ab)q(out)f(a)g(w)o(orkload)f(c)o (haracterization)h(study)h(at)f(a)f(national)g(sup)q(ercomputer)-90 954 y(cen)o(ter.)26 b(This)15 b(study)i(is)e(part)h(of)g(a)f(larger)h(pro)r (ject)h(to)f(dev)o(elop)f(an)h(analytic)f(metho)q(dology)f(for)h(b)q(enc)o (hmark)h(set)g(construction.)-90 1053 y(Our)j(approac)o(h)f(is)g(to)g(study)h (a)f(particular)g(w)o(orkload)f(in)h(detail)f(and)h(to)g(measure)h(its)f(p)q (erformance)g(c)o(haracteristics.)33 b(These)-90 1153 y(results)16 b(will)e(giv)o(e)g(us)i(a)e(p)q(erformance)h(sp)q(eci\014cation)h(that)f(a)g (b)q(enc)o(hmark)f(set)i(represen)o(ting)h(this)e(w)o(orkload)f(m)o(ust)g (appro)o(ximate)-90 1253 y(in)j(a)g(measurable)f(w)o(a)o(y)m(.)28 b(W)m(e)16 b(can)i(then)g(in)o(v)o(estigate)f(ho)o(w)g(to)g(analytically)e (select)k(b)q(enc)o(hmarks)e(to)g(represen)o(t)j(this)d(particular)-90 1352 y(w)o(orkload.)k(W)m(e)15 b(exp)q(ect)i(the)f(metho)q(dology)d(w)o(e)i (dev)o(elop)h(to)f(b)q(e)h(applicable)e(to)h(w)o(orkloads)g(at)g(other)h (sites.)23 b(Our)16 b(study)g(rep)q(orts)-90 1452 y(on)c(292,254)e(user)k (jobs)e(o)o(v)o(er)g(a)g(p)q(erio)q(d)h(of)f(13)f(mon)o(ths)h(represen)o (ting)h(ab)q(out)g(one)f(half)f(of)h(the)h(w)o(all)e(clo)q(c)o(k)h(time)f(a)o (v)n(ailable)f(on)i(the)h(4-)-90 1552 y(pro)q(cessor)i(CRA)m(Y)d(Y-MP)h(at)f (the)i(National)d(Cen)o(ter)j(for)f(Sup)q(ercomputing)f(Applications)g (during)h(this)f(time.)17 b(The)c(p)q(erformance)-90 1651 y(statistics)i(w)o (e)f(gathered)g(in)g(some)f(sense)i Fk(de\014ne)g Fl(the)g(w)o(orkload)d(b)q (ecause)k(of)d(the)h(large)g(fraction)f(of)h(the)g(a)o(v)n(ailable)e(time)g (that)i(w)o(as)-90 1751 y(recorded.)30 b(These)18 b(statistics)g(will)e(allo) o(w)g(us)h(to)h(\\rev)o(erse)g(engineer")g(a)f(set)i(of)d(b)q(enc)o(hmarks)h (that)h(analytically)d(represen)o(t)k(the)-90 1850 y(w)o(orkload)13 b(at)g(this)h(site.)p eop %%Page: 1 3 bop -90 195 a Fj(1)69 b(In)n(tro)r(duction)-90 336 y Fl(This)13 b(pap)q(er)h(describ)q(es)h(some)d(preliminary)f(results)j(ab)q(out)f(a)g(w)o (orkload)f(c)o(haracterization)h(study)h(at)f(a)f(national)g(sup)q (ercomputer)-90 435 y(cen)o(ter.)26 b(This)15 b(study)i(is)e(part)h(of)g(a)f (larger)h(pro)r(ject)h(to)f(dev)o(elop)f(an)h(analytic)f(metho)q(dology)f (for)h(b)q(enc)o(hmark)h(set)g(construction.)-90 535 y(Our)j(approac)o(h)f (is)g(to)g(study)h(a)f(particular)g(w)o(orkload)f(in)h(detail)f(and)h(to)g (measure)h(its)f(p)q(erformance)g(c)o(haracteristics.)33 b(These)-90 635 y(results)16 b(will)e(giv)o(e)g(us)i(a)e(p)q(erformance)h(sp)q (eci\014cation)h(that)f(a)g(b)q(enc)o(hmark)f(set)i(represen)o(ting)h(this)e (w)o(orkload)f(m)o(ust)g(appro)o(ximate)-90 734 y(in)j(a)g(measurable)f(w)o (a)o(y)m(.)28 b(W)m(e)16 b(can)i(then)g(in)o(v)o(estigate)f(ho)o(w)g(to)g (analytically)e(select)k(b)q(enc)o(hmarks)e(to)g(represen)o(t)j(this)d (particular)-90 834 y(w)o(orkload.)g(W)m(e)c(exp)q(ect)j(the)e(metho)q (dology)e(w)o(e)i(dev)o(elop)g(to)g(b)q(e)g(applicable)f(to)h(w)o(orkloads)f (at)h(other)g(sites.)-28 934 y(W)m(e)i(are)h(motiv)n(ated)e(b)o(y)h(a)h (curren)o(t)h(lac)o(k)e(of)g(detailed)g(understanding)h(\(quan)o (ti\014cation\))f(ab)q(out)h(ho)o(w)f(computers)h(are)g(used.)-90 1033 y(There)12 b(is)g(a)f(need)h(to)f(kno)o(w)g(more)f(ab)q(out)i(the)f(use) i(of)d(computers)i(than)f(the)h(n)o(um)o(b)q(er)f(of)f(hours)i(consumed)f(in) g(di\013eren)o(t)h(application)-90 1133 y(areas[20].)k(The)10 b(limited)e(p)q(erformance)i(information)d(a)o(v)n(ailable)h(ab)q(out)i(ho)o (w)g(computers)g(are)g(used)h(mak)o(es)e(it)h(di\016cult)f(to)h(construct)-90 1232 y(b)q(enc)o(hmarks)15 b(that)h(are)g(represen)o(tativ)o(e)h(of)e(the)i (w)o(orkloads)d(at)i(v)n(arious)e(sites)j(or)e(of)g(sup)q(ercomputers)i(in)e (general,)h(if)f(w)o(orkloads)-90 1332 y(are)f(not)g(kno)o(wn)g(or)f (di\016cult)h(to)f(understand)i(in)f(a)f(quan)o(ti\014able)g(w)o(a)o(y)m(.) -28 1432 y(W)m(e)i(b)q(eliev)o(e)h(that)f(w)o(e)g(m)o(ust)f(address)j(some)d (basic)h(problems)f(in)h(p)q(erformance)g(ev)n(aluation)f(that)h(are)h(still) e(as)h(true)h(to)q(da)o(y)f(as)-90 1531 y(they)h(w)o(ere)h(20)f(y)o(ears)g (ago[18)n(]:)22 b(Grenander)16 b(and)g(Tsao\(1971\))f(wrote:)23 b(\\W)m(e)15 b(b)q(eliev)o(e)h(that)g(no)g(real)g(signi\014can)o(t)f(adv)n (ance)h(in)f(the)-90 1631 y(ev)n(aluation)c(of)h(systems)h(can)g(b)q(e)g(exp) q(ected)h(un)o(til)e(some)g(breakthrough)h(is)f(made)g(in)g(the)h(c)o (haracterization)g(of)f(the)h(w)o(orkload[11)n(].")-90 1731 y(F)m(errari\(1972\))i(states:)24 b(\\The)16 b(lac)o(k)f(of)g(satisfactory)h (w)o(orkload)f(c)o(haracterization)i(tec)o(hniques)g(is)f(one)g(of)f(the)i (main)c(reasons)k(for)-90 1830 y(the)d(primitiv)o(e)e(state)j(of)e(this)h (imp)q(ortan)o(t)e(branc)o(h)j(of)e(computer)g(engineering")h(\(i.e.,)f(p)q (erformance)h(ev)n(aluation\)[9)n(].)-28 1930 y(T)m(o)h(address)j(these)f (problems)e(at)h(least)g(three)h(approac)o(hes)g(to)e(w)o(orkload)g(c)o (haracterization)i(and)e(b)q(enc)o(hmark)h(construction)-90 2029 y(ha)o(v)o(e)e(b)q(een)h(prop)q(osed[10]:)-28 2112 y Fi(\017)21 b Fl(Resource)14 b(Consumption)e(Approac)o(h.)18 b(In)13 b(this)h(approac)o (h)f(a)g(b)q(enc)o(hmark)g(set)h(represen)o(ts)i(a)d(w)o(orkload)f(w)o(ell)h (if)f(it)h(consumes)14 2162 y(resources)j(at)e(the)g(same)f(rate)i(as)f(the)g (real)g(w)o(orkload,)e(e.g.)18 b(CPU)c(time)f(used,)h(memory)d(space,)k(and)f (duration)f(of)g(I/O.)-28 2242 y Fi(\017)21 b Fl(F)m(unctional)16 b(Approac)o(h.)27 b(In)17 b(this)g(approac)o(h)g(a)f(b)q(enc)o(hmark)h(set)g (represen)o(ts)j(a)d(w)o(orkload)e(w)o(ell)h(if)g(it)h(p)q(erforms)f(the)i (same)14 2292 y(functions)c(as)h(the)g(w)o(orkload,)f(e.g.)20 b(a)14 b(real)h(w)o(orkload)e(p)q(erforming)g(pa)o(yroll)g(activities)i (should)f(b)q(e)h(represen)o(ted)j(b)o(y)c(pa)o(yroll)14 2342 y(programs.)j(This)c(application)g(area)h(co)o(v)o(erage)g(approac)o(h)g(is)g (follo)o(w)o(ed)e(b)o(y)i(man)o(y)e(curren)o(t)j(b)q(enc)o(hmarks,)e(e.g.)18 b(P)o(erfect[1)q(].)-28 2421 y Fi(\017)j Fl(P)o(erformance)13 b(Orien)o(ted)i(Approac)o(h.)j(In)c(this)g(approac)o(h)f(a)h(b)q(enc)o(hmark) f(set)i(represen)o(ts)h(a)e(w)o(orkload)e(w)o(ell)h(if)g(it)g(causes)i(the)14 2471 y(system)h(to)h(exhibit)f(the)h(same)f(p)q(erformance)h(c)o (haracteristics)h(as)f(the)g(w)o(orkload,)f(e.g.)26 b(M\015ops,)17 b(a)o(v)o(erage)g(v)o(ector)g(length.)14 2521 y(This)d(approac)o(h)g(w)o(as)g (used)h(at)f(Los)h(Alamos)d(on)i(the)h(CRA)m(Y-1)e(b)o(y)h(soft)o(w)o(are)g (instrumen)o(tation)f(of)g(time-consuming)f(co)q(des)14 2571 y(\(without)h(the)i(use)g(of)e(a)h(hardw)o(are)g(p)q(erformance)g (monitor\)[6)m(][16)o(].)-28 2704 y(While)f(eac)o(h)h(approac)o(h)f(has)h (disadv)n(an)o(tages,)f(w)o(e)g(\014nd)h(the)g(P)o(erformance)f(Orien)o(ted)i (Approac)o(h)f(the)g(most)e(attractiv)o(e.)18 b(It)c(has)950 2828 y(1)p eop %%Page: 2 4 bop -90 195 a Fl(the)16 b(adv)n(an)o(tage)e(that)h(it)f(forces)i(the)f(c)o (haracterization)h(and)f(construction)h(pro)q(cess)g(to)f(fo)q(cus)g(on)g (the)h(p)q(erformance)e(v)n(ariables)h(of)-90 295 y(in)o(terest)g(in)f(an)o (y)f(particular)h(study)m(.)-28 394 y(Kno)o(wing)h(ho)o(w)g(computers)g(are)h (used)g(is)f(the)h(\014rst)h(step)f(in)f(constructing)h(b)q(enc)o(hmark)f (sets)i(that)e(represen)o(ts)j(the)e(w)o(orkload)-90 494 y(of)e(those)g (computers.)19 b(If)14 b(one)g(w)o(ere)h(to)f(construct)i(a)e(b)q(enc)o (hmark)g(set)h(to)f(represen)o(t)i(a)e(w)o(orkload)f(then)i(the)f(selected)i (b)q(enc)o(hmark)-90 594 y(programs)10 b(should)h(collectiv)o(ely)f(ha)o(v)o (e)h(a)g(distribution)g(of)f(p)q(erformance)h(c)o(haracteristics)i(that)e(is) g(similar)e(to)i(that)g(of)g(the)h(w)o(orkload.)-90 693 y(In)18 b(this)g(pap)q(er)h(w)o(e)f(see,)h(for)f(the)h(\014rst)f(time,)f(what)h(the)h (v)n(alues)e(of)h(v)n(arious)f(p)q(erformance)h(metrics)g(actually)f(are)h (for)f(the)i(real)-90 793 y(w)o(orkload)13 b(at)g(an)h(academic)f(site.)19 b(W)m(e)13 b(ha)o(v)o(e)h(selected)i(the)e(academic)f(w)o(orkload)g(of)g(the) i(CRA)m(Y)e(Y-MP)h(at)g(the)g(National)f(Cen)o(ter)-90 892 y(for)g(Sup)q(ercomputing)h(Applications)f(\(NCSA\))h(as)g(our)g(initial)e (test)i(case.)19 b(This)14 b(selection)h(w)o(as)e(made)g(to)g(tak)o(e)h(adv)n (an)o(tage)f(of)g(the)-90 992 y(unique)e(opp)q(ortunit)o(y)g(a\013orded)h(b)o (y)f(the)g(installation)e(of)i(sp)q(ecial)g(recording)h(soft)o(w)o(are)f (that)g(has)h(collected)g(hardw)o(are)f(p)q(erformance)-90 1092 y(monitor\(HPM\))i(information)e(on)i(the)i(w)o(orkload)d(of)i(this)g (mac)o(hine)e(since)j(June)g(1991.)-28 1191 y(Other)d(studies)g(ha)o(v)o(e)f (b)q(een)h(made)e(using)h(the)h(HPM)f(\(primarily)e(on)i(the)g(CRA)m(Y)f (X-MP\))i(for)f(program)e(and)i(w)o(orkload)f(analysis)-90 1291 y(in)j(an)g(e\013ort)h(to)f(gain)f(a)h(more)g(detailed)g(understanding)h (of)f(the)g(p)q(erformance)h(c)o(haracteristics)h(of)d(b)q(enc)o(hmarks,)h (user)i(programs,)-90 1391 y(and)f(w)o(orkloads.)j(These)e(studies)g (include:)-28 1482 y Fi(\017)21 b Fl(Bradley[5)o(])14 b(-)f(5)h(co)q(des)h (from)d(the)j(P)o(erfect)g(Benc)o(hmark)-28 1565 y Fi(\017)21 b Fl(Berry[2])14 b(-)f(9)h(co)q(des)h(from)d(the)j(P)o(erfect)g(Benc)o(hmark) -28 1648 y Fi(\017)21 b Fl(Bradley[4)o(])14 b(-)f(11)h(co)q(des)h(from)d (large)i(time-consuming)d(users)k(at)f(NCSA)-28 1731 y Fi(\017)21 b Fl(Delic[8)o(])13 b(-)h(24)f(co)q(des)i(from)d(the)j(users)g(at)f(the)g (Ohio)g(Sup)q(ercomputer)h(Cen)o(ter)-28 1814 y Fi(\017)21 b Fl(Berry[3])14 b(-)f(most)g(of)g(the)i(curren)o(t,)g(p)q(opular)e(b)q(enc)o (hmark)h(set)-28 1897 y Fi(\017)21 b Fl(Nelson[17)o(])13 b(-)h(30)g(hours)g (at)g(the)g(La)o(wrence)h(Liv)o(ermore)e(National)f(Lab)q(oratory)-28 1980 y Fi(\017)21 b Fl(Sato[19)o(])13 b(-)h(8)f(da)o(ys)h(at)g(the)g (National)f(Cen)o(ter)i(for)e(A)o(tmospheric)h(Researc)o(h)-28 2063 y Fi(\017)21 b Fl(William)o(s[22)l(][23)o(])13 b(-)h(7)g(w)o(eek)o(ends) h(at)f(2)f(go)o(v)o(ernmen)o(t)g(sites)-28 2146 y Fi(\017)21 b Fl(W)m(est[24)o(])13 b(-)h(35)f(da)o(ys)h(at)g(the)g(On)o(tario)g(Cen)o (tre)h(for)e(Large)h(Scale)h(Computation.)-28 2287 y(Our)g(study)f(rep)q (orts)i(on)e(292,254)e(user)j(jobs)f(o)o(v)o(er)g(a)g(p)q(erio)q(d)g(of)g(13) f(mon)o(ths)g(represen)o(ting)j(ab)q(out)e(one-half)f(of)h(the)g(w)o(all)f (clo)q(c)o(k)-90 2387 y(time)d(a)o(v)n(ailable)g(on)h(the)i(4-pro)q(cessor)g (CRA)m(Y)e(Y-MP)h(at)f(NCSA)h(during)g(this)g(time.)k(The)c(p)q(erformance)f (statistics)i(w)o(e)f(gathered)g(in)-90 2487 y(some)g(sense)i Fk(de\014ne)f Fl(the)g(w)o(orkload)e(b)q(ecause)j(of)e(the)h(large)f (fraction)g(of)g(the)h(a)o(v)n(ailable)d(time)h(that)i(w)o(as)f(recorded.)19 b(These)14 b(statistics)-90 2586 y(will)e(allo)o(w)h(us)h(to)g(\\rev)o(erse)h (engineer")g(a)e(set)i(of)f(b)q(enc)o(hmarks)f(represen)o(ting)j(the)e(w)o (orkload)f(at)h(this)g(site.)950 2828 y(2)p eop %%Page: 3 5 bop -28 195 a Fl(The)17 b(next)f(section)h(b)q(egins)f(b)o(y)g(giving)e(an)i (o)o(v)o(erview)f(of)h(the)g(CRA)m(Y)f(Y-MP)h(arc)o(hitecture)i(and)e(its)g (hardw)o(are)g(p)q(erformance)-90 295 y(monitor.)23 b(The)17 b(system)f(soft)o(w)o(are)g(that)h(records)h(HPM)e(data)g(is)g(then)h (describ)q(ed.)27 b(The)17 b(metho)q(dology)d(section)j(ends)g(with)f(the)-90 394 y(de\014nitions)c(of)g(the)h(p)q(erformance)g(metrics)f(and)g(c)o (haracteristics)j(that)d(w)o(e)h(used)g(to)g(measure)f(the)h(w)o(orkload.)k (The)c(results)g(section)-90 494 y(follo)o(ws)i(next)i(and)f(giv)o(es)g (statistics)h(ab)q(out)g(the)g(w)o(orkload)e(as)i(a)f(whole)g(and)g(the)h (top)f(10)g(time-consuming)e(application)h(areas.)-90 594 y(Our)f (conclusions)h(and)e(plans)h(for)g(future)g(w)o(ork)g(are)g(found)f(in)h(the) g(last)g(section.)-90 806 y Fj(2)69 b(Metho)r(dology)-90 946 y Fl(This)15 b(section)g(giv)o(es)g(an)g(brief)g(o)o(v)o(erview)f(of)h(the)g (CRA)m(Y)f(Y-MP)h(arc)o(hitecture)i(and)e(the)g(hardw)o(are)h(p)q(erformance) e(monitor.)20 b(The)-90 1046 y(system)13 b(soft)o(w)o(are)g(that)g(records)h (HPM)g(information)c(ab)q(out)i(user)i(jobs)f(is)g(also)f(describ)q(ed.)20 b(Finally)m(,)10 b(w)o(e)k(de\014ne)g(the)f(p)q(erformance)-90 1146 y(metrics)h(deriv)o(ed)g(from)e(the)j(HPM)f(coun)o(ters.)-90 1320 y Fh(2.1)56 b(CRA)-5 b(Y)20 b(Y-MP)f(arc)n(hitecture)e(and)i(the)g (Hardw)n(are)g(P)n(erformance)e(Monitor)-90 1446 y Fl(In)12 b(order)g(to)f(b)q(etter)i(appreciate)g(the)f(p)q(erformance)f(data)g(that)h (is)f(b)q(eing)h(collected)g(and)f(to)h(more)e(easily)h(in)o(terpret)i(what)e (it)g(means,)-90 1546 y(it)j(is)h(useful)g(to)g(review)g(the)g(CRA)m(Y)f (Y-MP)h(arc)o(hitecture)i(and)e(the)g(hardw)o(are)g(p)q(erformance)g (monitor.)k(A)c(picture)g(of)f(the)i(basic)-90 1646 y(arc)o(hitecture)f(of)d (a)h(single)g(CRA)m(Y)f(Y-MP)h(pro)q(cessor)i(is)e(giv)o(en)f(in)h(Figure)g (1.)18 b(The)13 b(memory)e(p)q(orts,)i(register)h(sets,)g(and)f(functional) -90 1745 y(units)h(are)g(illustrated)g(individually)m(.)h(A)f(clo)q(c)o(k)g (p)q(erio)q(d)g(on)g(the)g(CRA)m(Y)f(Y-MP)i(is)e(6)h(nanoseconds.)-28 1845 y(Mac)o(hine)f(instructions)h(con)o(tained)e(in)h(the)g(instruction)g (bu\013ers)h(are)f(deco)q(ded)h(and)f(signals)f(are)h(generated)h(to)f(the)g (rest)h(of)e(the)-90 1944 y(CPU.)h(These)h(signals)f(are)g(issued)h(\(i.e.)j (the)d(instruction)f(issues\))i(when,)e(in)f(general,)h(the)h(resources)h (\(registers,)g(functional)d(units)-90 2044 y(or)j(p)q(orts\))g(needed)h(b)o (y)f(the)g(instruction)g(are)g(not)g(reserv)o(ed)h(or)f(in)f(use)i(b)o(y)e (some)g(previous)h(instruction.)21 b(Once)16 b(an)e(instruction)h(is)-90 2144 y(issued,)h(the)h(instruction)e(functional)g(unit)g(attempts)h(to)f (issue)h(the)h(next)f(instruction,)g(and)f(do)q(es)h(not)g(w)o(ait)f(for)g (the)h(previously)-90 2243 y(issued)g(instruction)g(to)g(complete)f(its)g (execution.)24 b(Hence,)17 b(man)o(y)d(instructions)i(ma)o(y)e(b)q(e)i(in)f (execution)i(sim)o(ultaneously)m(.)j(If)15 b(the)-90 2343 y(resources)i (needed)f(b)o(y)f(an)f(instruction)h(are)g(curren)o(tly)h(not)e(a)o(v)n (ailable,)f(then)i(the)g(instruction)g(functional)f(unit)g(\\holds)g(issue")i (of)-90 2443 y(the)c(instruction)f(un)o(til)g(the)g(needed)i(resources)h(b)q (ecome)d(a)o(v)n(ailable.)j(Note)e(that)f(while)g(the)h(instruction)f (functional)f(unit)h(is)g(holding)-90 2542 y(issue,)j(the)h(rest)g(of)e(the)h (CPU)h(is)e(busy)i(executing)f(previously)g(issued)h(instructions.)-28 2642 y(The)h(CPU)f(logic)f(that)h(deco)q(des)i(and)e(issues)h(instructions)g (for)e(execution)i(ma)o(y)d(b)q(e)j(considered)g(an)f(instruction)g (\\functional)950 2828 y(3)p eop %%Page: 4 6 bop 265 498 a Fm(Memory)832 499 y(Registers)1162 503 y(F)l(unctional)13 b(Units)p 244 2289 230 2 v 244 549 V 244 2289 2 1742 v 472 2289 V 277 1427 a Fl(Cen)o(tral)277 1490 y(Memory)p 688 2289 38 2 v 688 549 V 688 2289 2 1742 v 724 2289 V 496 2289 110 2 v 496 2145 V 496 2289 2 146 v 604 2289 V 511 2197 a(I/O)511 2256 y(\(FU\))493 1166 y(Memory)493 1220 y(P)o(orts)493 1268 y(\(FUs\))610 1378 y Fm(A)468 1389 y Fg(.)-6 b(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.) f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)f(.)683 1390 y(.)e(.)g(.)h(.)f(.)674 1391 y(.)h(.)671 1392 y(.)g(.)668 1393 y(.)f(.)665 1394 y(.)g(.)662 1395 y(.)660 1396 y(.)h(.)657 1397 y(.)656 1398 y(.)654 1399 y(.)g(.)652 1400 y(.)684 1389 y(.)g(.)f(.)g(.)h(.)676 1388 y(.)f(.)h(.)671 1387 y(.)g(.)668 1386 y(.)f(.)665 1385 y(.)663 1384 y(.)h(.)660 1383 y(.)659 1382 y(.)f(.)656 1381 y(.)654 1380 y(.)653 1379 y(.)h(.)611 1438 y Fm(B)468 1449 y Fg(.)j(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)f(.)683 1450 y(.)e(.)g(.)h(.)f(.)674 1451 y(.)h(.)671 1452 y(.)g(.)668 1453 y(.)f(.)665 1454 y(.)g(.)662 1455 y(.)660 1456 y(.)h(.)657 1457 y(.)656 1458 y(.)654 1459 y(.)g(.)652 1460 y(.)684 1449 y(.)g(.)f(.)g(.)h(.)676 1448 y(.)f(.)h(.)671 1447 y(.)g(.)668 1446 y(.)f(.)665 1445 y(.)663 1444 y(.)h(.)660 1443 y(.)659 1442 y(.)f(.)656 1441 y(.)654 1440 y(.)653 1439 y(.)h(.)611 1498 y Fm(C)684 1509 y Fg(.)g(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h (.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f (.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)h(.)f(.)g(.)h(.)f(.)g(.)h (.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f (.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g (.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)h(.)f (.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g (.)h(.)f(.)g(.)i(.)i(.)g(.)f(.)h(.)477 1508 y(.)f(.)h(.)481 1507 y(.)g(.)485 1506 y(.)f(.)488 1505 y(.)489 1504 y(.)h(.)492 1503 y(.)494 1502 y(.)f(.)497 1501 y(.)498 1500 y(.)500 1499 y(.)g(.)468 1509 y(.)470 1510 y(.)h(.)f(.)h(.)g(.)478 1511 y(.)g(.)481 1512 y(.)g(.)485 1513 y(.)f(.)488 1514 y(.)g(.)491 1515 y(.)492 1516 y(.)h(.)495 1517 y(.)497 1518 y(.)498 1519 y(.)g(.)501 1520 y(.)654 2265 y(.)654 2264 y(.)654 2262 y(.)654 2260 y(.)654 2259 y(.)654 2257 y(.)654 2255 y(.)654 2254 y(.)654 2252 y(.)654 2250 y(.)654 2249 y(.)654 2247 y(.)654 2245 y(.)654 2244 y(.)654 2242 y(.)654 2240 y(.)654 2239 y(.)654 2237 y(.)654 2235 y(.)654 2234 y(.)654 2232 y(.)654 2230 y(.)654 2228 y(.)654 2227 y(.)654 2225 y(.)654 2223 y(.)654 2222 y(.)654 2220 y(.)654 2218 y(.)654 2217 y(.)654 2215 y(.)654 2213 y(.)654 2212 y(.)654 2210 y(.)654 2208 y(.)654 2207 y(.)654 2205 y(.)654 2203 y(.)654 2202 y(.)654 2200 y(.)654 2198 y(.)654 2197 y(.)654 2195 y(.)654 2193 y(.)654 2192 y(.)654 2190 y(.)654 2188 y(.)654 2187 y(.)654 2185 y(.)654 2183 y(.)654 2182 y(.)654 2180 y(.)654 2178 y(.)654 2176 y(.)654 2175 y(.)654 2173 y(.)654 2171 y(.)654 2170 y(.)654 2168 y(.)654 2166 y(.)654 2165 y(.)654 2163 y(.)654 2161 y(.)654 2160 y(.)654 2158 y(.)654 2156 y(.)654 2155 y(.)654 2153 y(.)654 2151 y(.)654 2150 y(.)654 2148 y(.)654 2146 y(.)654 2145 y(.)654 2143 y(.)654 2141 y(.)654 2140 y(.)654 2138 y(.)654 2136 y(.)654 2135 y(.)654 2133 y(.)654 2131 y(.)654 2130 y(.)654 2128 y(.)654 2126 y(.)654 2124 y(.)654 2123 y(.)654 2121 y(.)654 2119 y(.)654 2118 y(.)654 2116 y(.)654 2114 y(.)654 2113 y(.)654 2111 y(.)654 2109 y(.)e(.)654 2111 y(.)655 2113 y(.)655 2114 y(.)655 2116 y(.)655 2118 y(.)656 2119 y(.)656 2121 y(.)657 2122 y(.)657 2124 y(.)657 2126 y(.)658 2127 y(.)659 2129 y(.)659 2130 y(.)660 2132 y(.)660 2133 y(.)661 2135 y(.)662 2136 y(.)663 2138 y(.)663 2139 y(.)664 2141 y(.)665 2142 y(.)654 2109 y(.)654 2111 y(.)654 2113 y(.)654 2114 y(.)653 2116 y(.)653 2118 y(.)653 2119 y(.)652 2121 y(.)652 2122 y(.)652 2124 y(.)651 2126 y(.)651 2127 y(.)650 2129 y(.)649 2130 y(.)649 2132 y(.)648 2133 y(.)647 2135 y(.)647 2136 y(.)646 2138 y(.)645 2139 y(.)644 2141 y(.)643 2142 y(.)528 1569 y(.)i(.)g(.)533 1568 y(.)534 1567 y(.)536 1566 y(.)537 1565 y(.)538 1563 y(.)539 1562 y(.)540 1561 y(.)540 1559 y(.)541 1558 y(.)542 1556 y(.)543 1555 y(.)543 1553 y(.)544 1552 y(.)545 1550 y(.)545 1549 y(.)546 1547 y(.)546 1545 y(.)547 1544 y(.)547 1542 y(.)548 1541 y(.)548 1539 y(.)549 1537 y(.)549 1536 y(.)549 1534 y(.)550 1533 y(.)550 1531 y(.)551 1529 y(.)551 1528 y(.)551 1526 y(.)552 1525 y(.)552 1523 y(.)552 1521 y(.)553 1520 y(.)553 1518 y(.)553 1516 y(.)553 1515 y(.)554 1513 y(.)554 1511 y(.)554 1510 y(.)554 1508 y(.)555 1507 y(.)555 1505 y(.)555 1503 y(.)555 1502 y(.)555 1500 y(.)556 1498 y(.)556 1497 y(.)556 1495 y(.)556 1493 y(.)556 1492 y(.)557 1490 y(.)557 1488 y(.)557 1487 y(.)557 1485 y(.)557 1483 y(.)557 1482 y(.)557 1480 y(.)557 1478 y(.)557 1477 y(.)558 1475 y(.)558 1473 y(.)558 1472 y(.)558 1470 y(.)558 1468 y(.)558 1467 y(.)558 1465 y(.)558 1463 y(.)558 1462 y(.)558 1460 y(.)558 1459 y(.)558 1457 y(.)558 1455 y(.)558 1454 y(.)558 1452 y(.)558 1450 y(.)558 1449 y(.)558 1447 y(.)558 1445 y(.)558 1444 y(.)558 1442 y(.)558 1440 y(.)558 1439 y(.)558 1437 y(.)558 1435 y(.)558 1434 y(.)558 1432 y(.)558 1430 y(.)558 1429 y(.)558 1427 y(.)558 1425 y(.)558 1424 y(.)557 1422 y(.)557 1420 y(.)557 1419 y(.)557 1417 y(.)557 1415 y(.)557 1414 y(.)557 1412 y(.)557 1410 y(.)557 1409 y(.)556 1407 y(.)556 1405 y(.)556 1404 y(.)556 1402 y(.)556 1400 y(.)555 1399 y(.)555 1397 y(.)555 1396 y(.)555 1394 y(.)555 1392 y(.)554 1391 y(.)554 1389 y(.)554 1387 y(.)554 1386 y(.)553 1384 y(.)553 1382 y(.)553 1381 y(.)553 1379 y(.)552 1377 y(.)552 1376 y(.)552 1374 y(.)551 1373 y(.)551 1371 y(.)551 1369 y(.)550 1368 y(.)550 1366 y(.)549 1365 y(.)549 1363 y(.)549 1361 y(.)548 1360 y(.)548 1358 y(.)547 1357 y(.)547 1355 y(.)546 1353 y(.)546 1352 y(.)545 1350 y(.)545 1349 y(.)544 1347 y(.)543 1346 y(.)543 1344 y(.)542 1343 y(.)541 1341 y(.)540 1340 y(.)540 1338 y(.)539 1337 y(.)538 1335 y(.)537 1334 y(.)536 1333 y(.)534 1332 y(.)533 1331 y(.)531 1330 y(.)d(.)528 1329 y(.)527 1330 y(.)f(.)524 1331 y(.)522 1332 y(.)521 1333 y(.)520 1334 y(.)519 1335 y(.)518 1337 y(.)517 1338 y(.)516 1340 y(.)515 1341 y(.)515 1343 y(.)514 1344 y(.)513 1346 y(.)513 1347 y(.)512 1349 y(.)511 1350 y(.)511 1352 y(.)510 1353 y(.)510 1355 y(.)509 1357 y(.)509 1358 y(.)508 1360 y(.)508 1361 y(.)507 1363 y(.)507 1365 y(.)507 1366 y(.)506 1368 y(.)506 1369 y(.)506 1371 y(.)505 1373 y(.)505 1374 y(.)505 1376 y(.)504 1378 y(.)504 1379 y(.)504 1381 y(.)503 1382 y(.)503 1384 y(.)503 1386 y(.)503 1387 y(.)502 1389 y(.)502 1391 y(.)502 1392 y(.)502 1394 y(.)501 1396 y(.)501 1397 y(.)501 1399 y(.)501 1401 y(.)501 1402 y(.)501 1404 y(.)500 1405 y(.)500 1407 y(.)500 1409 y(.)500 1410 y(.)500 1412 y(.)500 1414 y(.)500 1415 y(.)499 1417 y(.)499 1419 y(.)499 1420 y(.)499 1422 y(.)499 1424 y(.)499 1425 y(.)499 1427 y(.)499 1429 y(.)499 1430 y(.)499 1432 y(.)499 1434 y(.)498 1435 y(.)498 1437 y(.)498 1439 y(.)498 1440 y(.)498 1442 y(.)498 1444 y(.)498 1445 y(.)498 1447 y(.)498 1449 y(.)498 1450 y(.)498 1452 y(.)498 1454 y(.)498 1455 y(.)498 1457 y(.)498 1459 y(.)498 1460 y(.)498 1462 y(.)498 1464 y(.)499 1465 y(.)499 1467 y(.)499 1469 y(.)499 1470 y(.)499 1472 y(.)499 1473 y(.)499 1475 y(.)499 1477 y(.)499 1478 y(.)499 1480 y(.)499 1482 y(.)500 1483 y(.)500 1485 y(.)500 1487 y(.)500 1488 y(.)500 1490 y(.)500 1492 y(.)500 1493 y(.)501 1495 y(.)501 1497 y(.)501 1498 y(.)501 1500 y(.)501 1502 y(.)501 1503 y(.)502 1505 y(.)502 1507 y(.)502 1508 y(.)502 1510 y(.)503 1511 y(.)503 1513 y(.)503 1515 y(.)503 1516 y(.)504 1518 y(.)504 1520 y(.)504 1521 y(.)505 1523 y(.)505 1525 y(.)505 1526 y(.)506 1528 y(.)506 1529 y(.)506 1531 y(.)507 1533 y(.)507 1534 y(.)507 1536 y(.)508 1537 y(.)508 1539 y(.)509 1541 y(.)509 1542 y(.)510 1544 y(.)510 1545 y(.)511 1547 y(.)511 1549 y(.)512 1550 y(.)513 1552 y(.)513 1553 y(.)514 1555 y(.)515 1556 y(.)515 1558 y(.)516 1559 y(.)517 1561 y(.)518 1562 y(.)519 1563 y(.)520 1565 y(.)521 1566 y(.)522 1567 y(.)524 1568 y(.)525 1569 y(.)k(.)f(.)528 1571 y(.)528 1573 y(.)528 1574 y(.)528 1576 y(.)528 1578 y(.)528 1579 y(.)528 1581 y(.)528 1583 y(.)528 1584 y(.)528 1586 y(.)528 1588 y(.)528 1589 y(.)528 1591 y(.)528 1593 y(.)528 1594 y(.)528 1596 y(.)528 1598 y(.)528 1599 y(.)528 1601 y(.)528 1603 y(.)528 1604 y(.)528 1606 y(.)528 1608 y(.)528 1609 y(.)528 1611 y(.)528 1613 y(.)528 1614 y(.)528 1616 y(.)528 1618 y(.)528 1619 y(.)528 1621 y(.)528 1623 y(.)528 1624 y(.)528 1626 y(.)528 1628 y(.)528 1629 y(.)f(.)528 1628 y(.)528 1626 y(.)528 1624 y(.)527 1623 y(.)527 1621 y(.)527 1619 y(.)526 1618 y(.)526 1616 y(.)526 1615 y(.)525 1613 y(.)525 1612 y(.)524 1610 y(.)523 1608 y(.)523 1607 y(.)522 1605 y(.)521 1604 y(.)521 1602 y(.)520 1601 y(.)519 1600 y(.)518 1598 y(.)517 1597 y(.)528 1629 y(.)528 1628 y(.)529 1626 y(.)529 1624 y(.)529 1623 y(.)529 1621 y(.)530 1619 y(.)530 1618 y(.)531 1616 y(.)531 1615 y(.)531 1613 y(.)532 1612 y(.)533 1610 y(.)533 1608 y(.)534 1607 y(.)534 1605 y(.)535 1604 y(.)536 1602 y(.)537 1601 y(.)537 1600 y(.)538 1598 y(.)539 1597 y(.)495 1677 y Fm(ctr-6)543 2061 y(ctr-7)468 1809 y Fg(.)i(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.) h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)f(.)683 1810 y(.)e(.)g(.)h(.)f(.)674 1811 y(.)h(.)671 1812 y(.)g(.)668 1813 y(.)f(.)665 1814 y(.)g(.)662 1815 y(.)660 1816 y(.)h(.)657 1817 y(.)656 1818 y(.)654 1819 y(.)g(.)652 1820 y(.)684 1809 y(.)g(.)f(.)g(.)h(.)676 1808 y(.)f(.)h(.)671 1807 y(.)g(.)668 1806 y(.)f(.)665 1805 y(.)663 1804 y(.)h(.)660 1803 y(.)659 1802 y(.)f(.)656 1801 y(.)654 1800 y(.)653 1799 y(.)h(.)684 1809 y(.)g(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.) g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h (.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g (.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h (.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f (.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)h(.)f(.)g(.)h (.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f(.)g(.)h(.)f (.)g(.)i(.)i(.)g(.)f(.)h(.)477 1808 y(.)f(.)h(.)481 1807 y(.)g(.)485 1806 y(.)f(.)488 1805 y(.)489 1804 y(.)h(.)492 1803 y(.)494 1802 y(.)f(.)497 1801 y(.)498 1800 y(.)500 1799 y(.)g(.)468 1809 y(.)470 1810 y(.)h(.)f(.)h(.)g(.)478 1811 y(.)g(.)481 1812 y(.)g(.)485 1813 y(.)f(.)488 1814 y(.)g(.)491 1815 y(.)492 1816 y(.)h(.)495 1817 y(.)497 1818 y(.)498 1819 y(.)g(.)501 1820 y(.)562 1798 y Fm(D)p 808 837 290 2 v 808 741 V 808 837 2 98 v 1096 837 V 824 798 a Fl(V)13 b(Registers)p 808 1257 290 2 v 808 1161 V 808 1257 2 98 v 1096 1257 V 826 1218 a(S/T)h(Registers)p 808 1677 290 2 v 808 1581 V 808 1677 2 98 v 1096 1677 V 822 1638 a(A/B)h(Registers)p 808 2109 290 2 v 808 1989 V 808 2109 2 122 v 1096 2109 V 829 2039 a(Instruction)829 2085 y(Bu\013ers)600 2265 y Fg(.)-6 b(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)720 789 y(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)280 b(.)-6 b(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)g(.)f(.)1128 669 y(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)1128 1029 y(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.) h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)-44 b(.)1128 1028 y(.)1128 1026 y(.)1128 1024 y(.)1128 1023 y(.)1128 1021 y(.)1128 1019 y(.)1128 1018 y(.)1128 1016 y(.)1128 1014 y(.)1128 1013 y(.)1128 1011 y(.)1128 1009 y(.)1128 1008 y(.)1128 1006 y(.)1128 1004 y(.)1128 1003 y(.)1128 1001 y(.)1128 999 y(.)1128 998 y(.)1128 996 y(.)1128 994 y(.)1128 993 y(.)1128 991 y(.)1128 989 y(.)1128 988 y(.)1128 986 y(.)1128 984 y(.)1128 983 y(.)1128 981 y(.)1128 979 y(.)1128 978 y(.)1128 976 y(.)1128 974 y(.)1128 973 y(.)1128 971 y(.)1128 969 y(.)1128 968 y(.)1128 966 y(.)1128 964 y(.)1128 963 y(.)1128 961 y(.)1128 959 y(.)1128 958 y(.)1128 956 y(.)1128 954 y(.)1128 953 y(.)1128 951 y(.)1128 949 y(.)1128 948 y(.)1128 946 y(.)1128 944 y(.)1128 943 y(.)1128 941 y(.)1128 939 y(.)1128 938 y(.)1128 936 y(.)1128 934 y(.)1128 933 y(.)1128 931 y(.)1128 929 y(.)1128 928 y(.)1128 926 y(.)1128 924 y(.)1128 923 y(.)1128 921 y(.)1128 919 y(.)1128 918 y(.)1128 916 y(.)1128 914 y(.)1128 913 y(.)1128 911 y(.)1128 909 y(.)1128 908 y(.)1128 906 y(.)1128 904 y(.)1128 903 y(.)1128 901 y(.)1128 899 y(.)1128 898 y(.)1128 896 y(.)1128 894 y(.)1128 893 y(.)1128 891 y(.)1128 889 y(.)1128 888 y(.)1128 886 y(.)1128 884 y(.)1128 883 y(.)1128 881 y(.)1128 879 y(.)1128 878 y(.)1128 876 y(.)1128 874 y(.)1128 873 y(.)1128 871 y(.)1128 869 y(.)1128 868 y(.)1128 866 y(.)1128 864 y(.)1128 863 y(.)1128 861 y(.)1128 859 y(.)1128 858 y(.)1128 856 y(.)1128 854 y(.)1128 853 y(.)1128 851 y(.)1128 849 y(.)1128 848 y(.)1128 846 y(.)1128 844 y(.)1128 843 y(.)1128 841 y(.)1128 839 y(.)1128 838 y(.)1128 836 y(.)1128 834 y(.)1128 833 y(.)1128 831 y(.)1128 829 y(.)1128 828 y(.)1128 826 y(.)1128 824 y(.)1128 823 y(.)1128 821 y(.)1128 819 y(.)1128 818 y(.)1128 816 y(.)1128 814 y(.)1128 813 y(.)1128 811 y(.)1128 809 y(.)1128 808 y(.)1128 806 y(.)1128 804 y(.)1128 803 y(.)1128 801 y(.)1128 799 y(.)1128 798 y(.)1128 796 y(.)1128 794 y(.)1128 793 y(.)1128 791 y(.)1128 789 y(.)1128 788 y(.)1128 786 y(.)1128 784 y(.)1128 783 y(.)1128 781 y(.)1128 779 y(.)1128 778 y(.)1128 776 y(.)1128 774 y(.)1128 773 y(.)1128 771 y(.)1128 769 y(.)1128 768 y(.)1128 766 y(.)1128 764 y(.)1128 763 y(.)1128 761 y(.)1128 759 y(.)1128 758 y(.)1128 756 y(.)1128 754 y(.)1128 753 y(.)1128 751 y(.)1128 749 y(.)1128 748 y(.)1128 746 y(.)1128 744 y(.)1128 743 y(.)1128 741 y(.)1128 739 y(.)1128 738 y(.)1128 736 y(.)1128 734 y(.)1128 733 y(.)1128 731 y(.)1128 729 y(.)1128 728 y(.)1128 726 y(.)1128 724 y(.)1128 723 y(.)1128 721 y(.)1128 719 y(.)1128 718 y(.)1128 716 y(.)1128 714 y(.)1128 713 y(.)1128 711 y(.)1128 709 y(.)1128 708 y(.)1128 706 y(.)1128 704 y(.)1128 703 y(.)1128 701 y(.)1128 699 y(.)1128 698 y(.)1128 696 y(.)1128 694 y(.)1128 693 y(.)1128 691 y(.)1128 689 y(.)1128 688 y(.)1128 686 y(.)1128 684 y(.)1128 683 y(.)1128 681 y(.)1128 679 y(.)1128 678 y(.)1128 676 y(.)1128 674 y(.)1128 673 y(.)1128 671 y(.)1128 669 y(.)1128 1089 y(.)-6 b(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g (.)f(.)720 1209 y(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)280 b(.)-6 b(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)g(.)f(.)1128 1377 y(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.) g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)-44 b(.)1128 1376 y(.)1128 1374 y(.)1128 1372 y(.)1128 1371 y(.)1128 1369 y(.)1128 1367 y(.)1128 1366 y(.)1128 1364 y(.)1128 1362 y(.)1128 1361 y(.)1128 1359 y(.)1128 1357 y(.)1128 1356 y(.)1128 1354 y(.)1128 1352 y(.)1128 1351 y(.)1128 1349 y(.)1128 1347 y(.)1128 1346 y(.)1128 1344 y(.)1128 1342 y(.)1128 1341 y(.)1128 1339 y(.)1128 1337 y(.)1128 1336 y(.)1128 1334 y(.)1128 1332 y(.)1128 1331 y(.)1128 1329 y(.)1128 1327 y(.)1128 1326 y(.)1128 1324 y(.)1128 1322 y(.)1128 1321 y(.)1128 1319 y(.)1128 1317 y(.)1128 1316 y(.)1128 1314 y(.)1128 1312 y(.)1128 1311 y(.)1128 1309 y(.)1128 1307 y(.)1128 1306 y(.)1128 1304 y(.)1128 1302 y(.)1128 1301 y(.)1128 1299 y(.)1128 1297 y(.)1128 1296 y(.)1128 1294 y(.)1128 1292 y(.)1128 1291 y(.)1128 1289 y(.)1128 1287 y(.)1128 1286 y(.)1128 1284 y(.)1128 1282 y(.)1128 1281 y(.)1128 1279 y(.)1128 1278 y(.)1128 1276 y(.)1128 1274 y(.)1128 1273 y(.)1128 1271 y(.)1128 1269 y(.)1128 1268 y(.)1128 1266 y(.)1128 1264 y(.)1128 1263 y(.)1128 1261 y(.)1128 1259 y(.)1128 1258 y(.)1128 1256 y(.)1128 1254 y(.)1128 1253 y(.)1128 1251 y(.)1128 1249 y(.)1128 1248 y(.)1128 1246 y(.)1128 1244 y(.)1128 1243 y(.)1128 1241 y(.)1128 1239 y(.)1128 1238 y(.)1128 1236 y(.)1128 1234 y(.)1128 1233 y(.)1128 1231 y(.)1128 1229 y(.)1128 1228 y(.)1128 1226 y(.)1128 1224 y(.)1128 1223 y(.)1128 1221 y(.)1128 1219 y(.)1128 1218 y(.)1128 1216 y(.)1128 1214 y(.)1128 1213 y(.)1128 1211 y(.)1128 1209 y(.)1128 1208 y(.)1128 1206 y(.)1128 1204 y(.)1128 1203 y(.)1128 1201 y(.)1128 1199 y(.)1128 1198 y(.)1128 1196 y(.)1128 1194 y(.)1128 1193 y(.)1128 1191 y(.)1128 1189 y(.)1128 1188 y(.)1128 1186 y(.)1128 1184 y(.)1128 1183 y(.)1128 1181 y(.)1128 1179 y(.)1128 1178 y(.)1128 1176 y(.)1128 1174 y(.)1128 1173 y(.)1128 1171 y(.)1128 1169 y(.)1128 1168 y(.)1128 1166 y(.)1128 1164 y(.)1128 1163 y(.)1128 1161 y(.)1128 1159 y(.)1128 1158 y(.)1128 1156 y(.)1128 1154 y(.)1128 1153 y(.)1128 1151 y(.)1128 1149 y(.)1128 1148 y(.)1128 1146 y(.)1128 1144 y(.)1128 1143 y(.)1128 1141 y(.)1128 1139 y(.)1128 1138 y(.)1128 1136 y(.)1128 1134 y(.)1128 1133 y(.)1128 1131 y(.)1128 1129 y(.)1128 1128 y(.)1128 1126 y(.)1128 1124 y(.)1128 1123 y(.)1128 1121 y(.)1128 1119 y(.)1128 1118 y(.)1128 1116 y(.)1128 1114 y(.)1128 1113 y(.)1128 1111 y(.)1128 1109 y(.)1128 1108 y(.)1128 1106 y(.)1128 1104 y(.)1128 1103 y(.)1128 1101 y(.)1128 1099 y(.)1128 1098 y(.)1128 1096 y(.)1128 1094 y(.)1128 1093 y(.)1128 1091 y(.)1128 1089 y(.)720 1629 y(.)-6 b(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.) g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)280 b(.)-6 b(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f (.)h(.)g(.)f(.)h(.)g(.)f(.)720 2049 y(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g(.)f(.)h (.)g(.)f(.)280 b(.)-6 b(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)1452 993 y(.)h(.)g(.)f(.)h(.)g(.)g (.)f(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)1452 1041 y(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.) h(.)g(.)f(.)h(.)g(.)g(.)f(.)1452 1089 y(.)h(.)g(.)f(.)h(.)g(.)g(.)f(.)h(.)g (.)f(.)h(.)g(.)g(.)f(.)f(.)1476 1088 y(.)1476 1086 y(.)1476 1084 y(.)1476 1083 y(.)1476 1081 y(.)1476 1079 y(.)1476 1078 y(.)1476 1076 y(.)1476 1074 y(.)1476 1073 y(.)1476 1071 y(.)1476 1069 y(.)1476 1068 y(.)1476 1066 y(.)1476 1064 y(.)1476 1063 y(.)1476 1061 y(.)1476 1059 y(.)1476 1058 y(.)1476 1056 y(.)1476 1054 y(.)1476 1053 y(.)1476 1051 y(.)1476 1049 y(.)1476 1048 y(.)1476 1046 y(.)1476 1044 y(.)1476 1043 y(.)1476 1041 y(.)1476 1039 y(.)1476 1037 y(.)1476 1036 y(.)1476 1034 y(.)1476 1032 y(.)1476 1031 y(.)1476 1029 y(.)1476 1027 y(.)1476 1026 y(.)1476 1024 y(.)1476 1022 y(.)1476 1021 y(.)1476 1019 y(.)1476 1017 y(.)1476 1016 y(.)1476 1014 y(.)1476 1012 y(.)1476 1011 y(.)1476 1009 y(.)1476 1007 y(.)1476 1006 y(.)1476 1004 y(.)1476 1002 y(.)1476 1001 y(.)1476 999 y(.)1476 997 y(.)1476 996 y(.)1476 994 y(.)1476 992 y(.)1476 991 y(.)1476 989 y(.)1476 987 y(.)1476 986 y(.)1476 984 y(.)1476 982 y(.)1476 981 y(.)1476 979 y(.)1476 977 y(.)1476 976 y(.)1476 974 y(.)1476 972 y(.)1476 971 y(.)1476 969 y(.)1476 967 y(.)1476 965 y(.)1476 964 y(.)1476 962 y(.)1476 960 y(.)1476 959 y(.)1476 957 y(.)1476 955 y(.)1476 954 y(.)1476 952 y(.)1476 950 y(.)1476 949 y(.)1476 947 y(.)1476 945 y(.)g(.)1476 947 y(.)1477 949 y(.)1477 950 y(.)1477 952 y(.)1477 954 y(.)1478 955 y(.)1478 957 y(.)1479 958 y(.)1479 960 y(.)1479 962 y(.)1480 963 y(.)1481 965 y(.)1481 966 y(.)1482 968 y(.)1482 969 y(.)1483 971 y(.)1484 972 y(.)1485 974 y(.)1485 975 y(.)1486 977 y(.)1487 978 y(.)1476 945 y(.)1476 947 y(.)1476 949 y(.)1476 950 y(.)1475 952 y(.)1475 954 y(.)1475 955 y(.)1474 957 y(.)1474 958 y(.)1474 960 y(.)1473 962 y(.)1473 963 y(.)1472 965 y(.)1471 966 y(.)1471 968 y(.)1470 969 y(.)1469 971 y(.)1469 972 y(.)1468 974 y(.)1467 975 y(.)1466 977 y(.)1465 978 y(.)1346 917 y Fm(ctr-3,4,5)768 2049 y Fg(.)768 2048 y(.)768 2046 y(.)768 2044 y(.)768 2043 y(.)768 2041 y(.)768 2039 y(.)768 2038 y(.)768 2036 y(.)768 2034 y(.)768 2033 y(.)768 2031 y(.)768 2029 y(.)768 2027 y(.)768 2026 y(.)768 2024 y(.)768 2022 y(.)768 2021 y(.)768 2019 y(.)768 2017 y(.)768 2016 y(.)768 2014 y(.)768 2012 y(.)768 2011 y(.)768 2009 y(.)768 2007 y(.)768 2006 y(.)768 2004 y(.)768 2002 y(.)768 2001 y(.)768 1999 y(.)768 1997 y(.)768 1995 y(.)768 1994 y(.)768 1992 y(.)768 1990 y(.)768 1989 y(.)768 1987 y(.)768 1985 y(.)768 1984 y(.)768 1982 y(.)768 1980 y(.)768 1979 y(.)768 1977 y(.)768 1975 y(.)768 1974 y(.)768 1972 y(.)768 1970 y(.)768 1969 y(.)768 1967 y(.)768 1965 y(.)768 1963 y(.)768 1962 y(.)768 1960 y(.)768 1958 y(.)768 1957 y(.)768 1955 y(.)768 1953 y(.)g(.)768 1955 y(.)769 1957 y(.)769 1958 y(.)769 1960 y(.)769 1962 y(.)770 1963 y(.)770 1965 y(.)771 1966 y(.)771 1968 y(.)771 1970 y(.)772 1971 y(.)773 1973 y(.)773 1974 y(.)774 1976 y(.)774 1977 y(.)775 1979 y(.)776 1980 y(.)777 1982 y(.)777 1983 y(.)778 1985 y(.)779 1986 y(.)768 1953 y(.)768 1955 y(.)768 1957 y(.)768 1958 y(.)767 1960 y(.)767 1962 y(.)767 1963 y(.)766 1965 y(.)766 1966 y(.)766 1968 y(.)765 1970 y(.)765 1971 y(.)764 1973 y(.)763 1974 y(.)763 1976 y(.)762 1977 y(.)761 1979 y(.)761 1980 y(.)760 1982 y(.)759 1983 y(.)758 1985 y(.)757 1986 y(.)735 1905 y Fm(ctr-2)p 1168 849 338 2 v 1168 549 V 1168 849 2 302 v 1504 849 V 1182 606 a Fl(In)o(terger)15 b(Add\(64\))1182 662 y(Logical)1182 716 y(Shift)1182 764 y(P)o(op/P)o(arit)o(y)1182 823 y(2nd/Logical)1554 718 y(V)m(ector)p 1168 1149 290 2 v 1168 969 V 1168 1149 2 182 v 1456 1149 V 1182 1013 a(FP)g(Add)1182 1058 y(FP)g(Multiply)1182 1112 y(FP)g(Recipro)q(cal)1525 1040 y(Floating)1525 1093 y(P)o(oin)o(t)p 1168 1497 338 2 v 1168 1257 V 1168 1497 2 242 v 1504 1497 V 1191 1307 a(In)o(teger)f(Add\(64\))1191 1364 y(Logical)1191 1417 y(Shift)1191 1466 y(P)o(op/P)o(arit)o(y/LZ)1557 1391 y(Scalar)p 1168 1701 338 2 v 1168 1581 V 1168 1701 2 122 v 1504 1701 V 1185 1621 a(In)o(teger)h(Add\(32\))1185 1680 y(In)o(teger)g(Mult\(32\))1516 1655 y(Address)p 1168 2169 290 2 v 1168 1989 V 1168 2169 2 182 v 1456 2169 V 1194 2037 a(Instruction)1194 2082 y(Deco)q(de)g(and)1194 2128 y(Issue)g(Logic)1308 1989 y Fg(.)1308 1988 y(.)1308 1986 y(.)1308 1984 y(.)1308 1983 y(.)1308 1981 y(.)1308 1979 y(.)1308 1978 y(.)1308 1976 y(.)1308 1974 y(.)1308 1973 y(.)1308 1971 y(.)1308 1969 y(.)1308 1968 y(.)1308 1966 y(.)1308 1964 y(.)1308 1963 y(.)1308 1961 y(.)1308 1959 y(.)1308 1958 y(.)1308 1956 y(.)1308 1954 y(.)1308 1953 y(.)1308 1951 y(.)1308 1949 y(.)1308 1948 y(.)1308 1946 y(.)1308 1944 y(.)1308 1943 y(.)1308 1941 y(.)1308 1939 y(.)1308 1938 y(.)1308 1936 y(.)1308 1934 y(.)1308 1933 y(.)1308 1931 y(.)1308 1929 y(.)1308 1928 y(.)1308 1926 y(.)1308 1924 y(.)1308 1923 y(.)1308 1921 y(.)1308 1919 y(.)1308 1918 y(.)1308 1916 y(.)1308 1914 y(.)1308 1913 y(.)1308 1911 y(.)1308 1909 y(.)1308 1908 y(.)1308 1906 y(.)1308 1904 y(.)1308 1903 y(.)1308 1901 y(.)1308 1899 y(.)1308 1898 y(.)1308 1896 y(.)1308 1894 y(.)1308 1893 y(.)1308 1891 y(.)1308 1889 y(.)1308 1888 y(.)1308 1886 y(.)1308 1884 y(.)1308 1883 y(.)1308 1881 y(.)1308 1879 y(.)1308 1878 y(.)1308 1876 y(.)1308 1874 y(.)1308 1873 y(.)1308 1871 y(.)1308 1869 y(.)-8 b(.)1308 1871 y(.)1309 1873 y(.)1309 1874 y(.)1309 1876 y(.)1309 1878 y(.)1310 1879 y(.)1310 1881 y(.)1311 1882 y(.)1311 1884 y(.)1311 1886 y(.)1312 1887 y(.)1313 1889 y(.)1313 1890 y(.)1314 1892 y(.)1314 1893 y(.)1315 1895 y(.)1316 1896 y(.)1317 1898 y(.)1317 1899 y(.)1318 1901 y(.)1319 1902 y(.)1308 1869 y(.)1308 1871 y(.)1308 1873 y(.)1308 1874 y(.)1307 1876 y(.)1307 1878 y(.)1307 1879 y(.)1306 1881 y(.)1306 1882 y(.)1306 1884 y(.)1305 1886 y(.)1305 1887 y(.)1304 1889 y(.)1303 1890 y(.)1303 1892 y(.)1302 1893 y(.)1301 1895 y(.)1301 1896 y(.)1300 1898 y(.)1299 1899 y(.)1298 1901 y(.)1297 1902 y(.)1239 1821 y Fm(ctr-0)1477 2086 y Fl(Instruction)1284 2229 y Fg(.)1284 2228 y(.)1284 2226 y(.)1284 2224 y(.)1284 2223 y(.)1284 2221 y(.)1283 2219 y(.)1283 2218 y(.)1283 2216 y(.)1282 2215 y(.)1282 2213 y(.)1282 2211 y(.)1281 2210 y(.)1280 2208 y(.)1280 2207 y(.)1279 2205 y(.)1278 2204 y(.)1278 2202 y(.)1277 2201 y(.)1276 2199 y(.)1275 2198 y(.)1274 2196 y(.)1273 2195 y(.)1273 2194 y(.)1272 2192 y(.)1270 2191 y(.)1269 2190 y(.)1268 2189 y(.)1267 2187 y(.)1266 2186 y(.)1265 2185 y(.)1264 2184 y(.)1262 2183 y(.)1261 2182 y(.)1260 2181 y(.)1258 2180 y(.)1257 2179 y(.)1255 2178 y(.)1254 2177 y(.)1253 2176 y(.)e(.)1250 2175 y(.)1248 2174 y(.)h(.)1245 2173 y(.)f(.)1242 2172 y(.)g(.)1239 2171 y(.)g(.)1235 2170 y(.)h(.)f(.)h(.)f(.)1227 2169 y(.)h(.)f(.)g(.)h(.)1219 2170 y(.)f(.)h(.)f(.)1212 2171 y(.)h(.)f(.)1207 2172 y(.)h(.)1204 2173 y(.)g(.)1201 2174 y(.)1200 2175 y(.)f(.)1197 2176 y(.)1195 2177 y(.)1194 2178 y(.)1192 2179 y(.)h(.)1190 2180 y(.)1188 2181 y(.)1187 2182 y(.)1186 2183 y(.)1184 2185 y(.)1183 2186 y(.)1182 2187 y(.)1181 2188 y(.)1180 2189 y(.)1179 2190 y(.)1178 2192 y(.)1177 2193 y(.)1176 2194 y(.)1175 2196 y(.)1174 2197 y(.)1173 2199 y(.)1172 2200 y(.)1171 2201 y(.)1170 2203 y(.)1170 2204 y(.)1169 2206 y(.)1168 2207 y(.)1168 2209 y(.)1167 2211 y(.)1167 2212 y(.)1166 2214 y(.)1166 2215 y(.)1166 2217 y(.)1165 2219 y(.)1165 2220 y(.)1165 2222 y(.)1165 2224 y(.)1164 2225 y(.)1164 2227 y(.)1164 2229 y(.)1164 2230 y(.)1164 2232 y(.)1164 2233 y(.)1165 2235 y(.)1165 2237 y(.)1165 2238 y(.)1165 2240 y(.)1166 2242 y(.)1166 2243 y(.)1166 2245 y(.)1167 2247 y(.)1167 2248 y(.)1168 2250 y(.)1168 2251 y(.)1169 2253 y(.)1170 2254 y(.)1170 2256 y(.)1171 2257 y(.)1172 2259 y(.)1173 2260 y(.)1174 2262 y(.)1175 2263 y(.)1176 2264 y(.)1177 2266 y(.)1178 2267 y(.)1179 2268 y(.)1180 2270 y(.)1181 2271 y(.)1182 2272 y(.)1183 2273 y(.)1184 2274 y(.)1186 2275 y(.)1187 2276 y(.)1188 2277 y(.)1190 2278 y(.)1191 2279 y(.)1192 2280 y(.)1194 2281 y(.)1195 2282 y(.)1197 2283 y(.)i(.)1200 2284 y(.)1201 2285 y(.)h(.)1204 2286 y(.)g(.)1207 2287 y(.)g(.)1211 2288 y(.)f(.)h(.)1216 2289 y(.)f(.)h(.)f(.)h(.)g(.)f(.)h(.)g(.)f(.)h(.)g(.)1235 2288 y(.)g(.)g(.)1240 2287 y(.)g(.)1243 2286 y(.)g(.)1247 2285 y(.)1248 2284 y(.)g(.)1251 2283 y(.)1253 2282 y(.)1254 2281 y(.)f(.)1257 2280 y(.)1258 2279 y(.)1260 2278 y(.)1261 2277 y(.)1262 2276 y(.)1263 2275 y(.)1265 2274 y(.)1266 2273 y(.)1267 2271 y(.)1268 2270 y(.)1269 2269 y(.)1270 2268 y(.)1271 2266 y(.)1272 2265 y(.)1273 2264 y(.)1274 2262 y(.)1275 2261 y(.)1276 2260 y(.)1277 2258 y(.)1278 2257 y(.)1278 2255 y(.)1279 2254 y(.)1280 2252 y(.)1280 2251 y(.)1281 2249 y(.)1281 2247 y(.)1282 2246 y(.)1282 2244 y(.)1283 2243 y(.)1283 2241 y(.)1283 2239 y(.)1284 2238 y(.)1284 2236 y(.)1284 2234 y(.)1284 2233 y(.)1284 2231 y(.)1284 2229 y(.)f(.)1284 2231 y(.)1284 2233 y(.)1284 2234 y(.)1284 2236 y(.)1284 2238 y(.)1284 2239 y(.)1284 2241 y(.)1284 2243 y(.)1284 2244 y(.)1284 2246 y(.)1284 2248 y(.)1284 2249 y(.)1284 2251 y(.)1284 2253 y(.)1284 2254 y(.)1284 2256 y(.)1284 2258 y(.)1284 2259 y(.)g(.)1284 2258 y(.)1284 2256 y(.)1284 2254 y(.)1283 2253 y(.)1283 2251 y(.)1283 2249 y(.)1282 2248 y(.)1282 2246 y(.)1282 2245 y(.)1281 2243 y(.)1281 2242 y(.)1280 2240 y(.)1279 2238 y(.)1279 2237 y(.)1278 2235 y(.)1277 2234 y(.)1277 2232 y(.)1276 2231 y(.)1275 2229 y(.)1274 2228 y(.)1273 2227 y(.)1284 2259 y(.)1284 2258 y(.)1285 2256 y(.)1285 2254 y(.)1285 2253 y(.)1285 2251 y(.)1286 2249 y(.)1286 2248 y(.)1287 2246 y(.)1287 2245 y(.)1287 2243 y(.)1288 2242 y(.)1289 2240 y(.)1289 2238 y(.)1290 2237 y(.)1290 2235 y(.)1291 2234 y(.)1292 2232 y(.)1293 2231 y(.)1293 2229 y(.)1294 2228 y(.)1295 2227 y(.)1359 2241 y Fm(ctr-1)245 2379 y Fl(Figure)14 b(1:)k(CRA)m(Y)13 b(Y-MP)h(Pro)q(cessor)i(Arc)o(hitecture)g (with)e(HPM)g(Group)f(0)h(Coun)o(ters.)950 2828 y(4)p eop %%Page: 5 7 bop -90 195 a Fl(unit".)22 b(The)16 b(\\op)q(erands")g(for)f(the)h (instruction)g(functional)e(unit)i(are)f(the)i(16-bit,)d(32-bit)g(and)i (48-bit)e(mac)o(hine)g(instructions)j(of)-90 295 y(the)c(user's)g(program.)i (The)e(\\results")f(are)h(the)f(signals)g(to)f(the)i(rest)g(of)f(the)g(CPU)g (that)g(cause)h(the)g(w)o(ork)f(sp)q(eci\014ed)h(in)f(the)g(mac)o(hine)-90 394 y(instruction)i(to)f(b)q(e)h(p)q(erformed.)j(The)d(16-bit)e(instructions) i(can)f(issue)h(in)f(1)g(clo)q(c)o(k)h(p)q(erio)q(d,)f(while)g(32-bit)f(and)h (48-bit)f(instructions)-90 494 y(can)i(issue)h(in)e(2)h(clo)q(c)o(k)g(p)q (erio)q(ds)g(if)f(no)h(hold)f(issue)i(conditions)e(exist.)19 b(The)14 b(execution)h(time)d(of)i(an)f(instruction)i(v)n(aries.)-28 594 y(There)20 b(are)e(4)g(instruction)g(bu\013ers)i(eac)o(h)e(of)g(whic)o(h) g(con)o(tain)g(32)f(w)o(ords)h(\(up)h(to)f(128)f(\(16-bit\))g(instruction)i (parcels\).)32 b(The)-90 693 y(instruction)17 b(bu\013ers)h(supply)e (instructions)i(to)e(the)h(instruction)g(deco)q(de)h(and)e(issue)h(logic)f (\(functional)g(unit\).)26 b(When)16 b(the)h(next)-90 793 y(instruction)c(to) g(b)q(e)g(executed)i(is)e(not)f(in)h(an)f(instruction)i(bu\013er,)f(an)g (instruction)g(bu\013er)h(fetc)o(h)f(from)e(memory)f(o)q(ccurs.)20 b(There)14 b(are)-90 892 y(4)i(p)q(orts)g(\(lab)q(eled)g(A,)g(B,)f(C,)h(and)f (D\))h(that)g(connect)h(eac)o(h)f(CPU)h(to)e(the)i(cen)o(tral)f(memory)m(.)21 b(P)o(orts)16 b(A)g(and)g(B)g(are)g(used)h(to)f(load)-90 992 y(op)q(erands)c(from)e(memory)f(and)i(p)q(ort)h(C)f(is)g(used)i(to)e(store)h (op)q(erands)h(to)e(memory)m(.)j(P)o(ort)e(D)f(is)g(used)h(to)f(fetc)o(h)h (program)e(instructions)-90 1092 y(and)16 b(to)g(p)q(erform)g(system)g(I/O.)g (A)h(job)e(running)i(on)f(a)g(pro)q(cessor)i(ma)o(y)c(use)k(p)q(ort)e(D)g(on) h(an)o(y)f(pro)q(cessor)i(in)e(the)h(system)f(for)g(its)-90 1191 y(I/O.)h(There)h(are)g(3)f(\015oating)f(p)q(oin)o(t)h(functional)f (units)h(that)h(p)q(erform)e(64-bit)g(\015oating)h(p)q(oin)o(t)f(op)q (erations:)25 b(addition,)17 b(m)o(ultiply)-90 1291 y(and)e(recipro)q(cal)h (appro)o(ximation.)j(Division)14 b(is)h(p)q(erformed)g(with)f(a)h(com)o (bination)e(of)i(recipro)q(cal)h(appro)o(ximatio)o(n)d(and)i(m)o(ultiply)-90 1391 y(op)q(erations.)-28 1490 y(Eac)o(h)g(CPU)g(of)f(a)h(CRA)m(Y)f(Y-MP)h (system)f(con)o(tains)h(a)f(set)i(of)e(8)g(p)q(erformance)h(monitor)e(coun)o (ters.)21 b(Eac)o(h)15 b(coun)o(ter)h(is)f(48)f(bits)-90 1590 y(in)h(length.)24 b(These)17 b(coun)o(ters)g(monitor)d(and)i(trac)o(k)g (certain)g(hardw)o(are)g(related)h(ev)o(en)o(ts)g(that)f(can)g(b)q(e)g(used)h (to)e(indicate)h(relativ)o(e)-90 1689 y(p)q(erformance.)30 b(Thirt)o(y-t)o(w)o(o)17 b(ev)o(en)o(ts)i(can)f(b)q(e)g(monitored)f(and)g (these)j(ev)o(en)o(ts)e(are)h(divided)e(in)o(to)g(four)h(groups:)26 b(0,)18 b(1,)g(2,)g(3.)30 b(A)o(t)-90 1789 y(an)o(y)17 b(one)g(time)e(only)i (one)g(group)g(of)f(8)h(ev)o(en)o(ts)h(is)f(asso)q(ciated)h(with)e(the)i(8)f (ph)o(ysical)f(hardw)o(are)i(coun)o(ters.)28 b(The)18 b(asso)q(ciation)f(of) -90 1889 y(an)d(ev)o(en)o(t)h(group)f(with)g(the)h(hardw)o(are)g(coun)o(ters) h(is)e(p)q(erformed)g(b)o(y)g(the)h(op)q(erating)f(system)h(soft)o(w)o(are.)k (P)o(erformance)14 b(ev)o(en)o(ts)i(are)-90 1988 y(monitored)e(only)h(while)g (the)h(CPU)g(is)f(in)g(user)i(mo)q(de.)22 b(During)15 b(eac)o(h)h(clo)q(c)o (k)f(p)q(erio)q(d)h(in)f(user)i(mo)q(de,)d(the)i(p)q(erformance)g(coun)o (ters)-90 2088 y(are)d(incremen)o(ted)f(according)h(to)f(the)h(n)o(um)o(b)q (er)e(of)h(monitored)f(ev)o(en)o(ts)i(that)g(o)q(ccur.)18 b(The)13 b(op)q(eration)f(of)g(the)h(p)q(erformance)f(monitor)-90 2188 y(coun)o(ters)17 b(is)f(done)g(in)f(parallel)g(with)g(the)h(execution)h(of)e (user)i(programs,)d(do)q(es)j(not)f(dela)o(y)f(or)h(in)o(terfere)g(with)g (the)g(ev)o(en)o(ts)h(b)q(eing)-90 2287 y(monitored,)9 b(and)i(is)f(part)h (of)f(the)h(normal)d(op)q(eration)j(of)f(the)h(mac)o(hine.)k(T)m(able)10 b(1)h(sho)o(ws)f(the)i(p)q(erformance)e(coun)o(ters)i(that)f(comprise)-90 2387 y(monitor)h(group)i(0.)k(Besides)d(the)g(8)e(coun)o(ter)i(v)n(alues,)e (the)i(accum)o(ulated)e(CPU)h(time)f(in)g(clo)q(c)o(k)h(p)q(erio)q(ds)h(is)f (also)f(recorded.)20 b(Figure)-90 2487 y(1)15 b(illustrates)h(the)g(parts)g (of)f(the)h(Y-MP)f(pro)q(cessor)j(arc)o(hitecture)f(that)f(are)f(monitored)g (b)o(y)g(the)h(group)f(0)g(coun)o(ters.)24 b(A)16 b(detailed)-90 2586 y(description)f(of)e(the)h(hardw)o(are)h(p)q(erformance)e(monitor)f (coun)o(ters)j(ma)o(y)d(b)q(e)j(found)e(in)h([21)o(][14)n(].)950 2828 y(5)p eop %%Page: 6 8 bop 29 197 1863 2 v 28 247 2 50 v 37 247 V 62 232 a Fl(P)o(erformance)p 313 247 V 50 w(Description)p 1650 247 V 1130 w(Incremen)o(t)p 1882 247 V 1891 247 V 28 296 V 37 296 V 62 281 a(Coun)o(ter)p 313 296 V 131 w(Num)o(b)q(er)13 b(of:)p 1650 296 V 1131 w(P)o(er)h(CP)p 1882 296 V 1891 296 V 29 298 1863 2 v 28 348 2 50 v 37 348 V 62 333 a(0)p 313 348 V 256 w Fm(Instruction)o(s)f(issued)p Fl(.)j(This)e(coun)o(ter)g(is)g(incremen)o(ted)g(b)o(y)g(1)g(when)g(an)p 1650 348 V 151 w(+1)p 1882 348 V 1891 348 V 28 398 V 37 398 V 313 398 V 339 383 a(instruction)g(is)g(issued.)p 1650 398 V 1882 398 V 1891 398 V 29 399 1863 2 v 28 449 2 50 v 37 449 V 62 434 a(1)p 313 449 V 256 w Fm(Clo)q(c)o(k)i(p)q(erio)q(ds)e(holding)e (issue)p Fl(.)17 b(This)c(coun)o(ter)i(is)f(incremen)o(ted)g(b)o(y)g(1)p 1650 449 V 137 w(+1)p 1882 449 V 1891 449 V 28 499 V 37 499 V 313 499 V 339 484 a(when)g(an)g(instruction)g(is)g(prev)o(en)o(ted)h(from)d (issuing.)p 1650 499 V 1882 499 V 1891 499 V 29 501 1863 2 v 28 550 2 50 v 37 550 V 62 535 a(2)p 313 550 V 256 w Fm(Instruction)g (bu\013er)i(fetc)o(hes)h(\(p)q(ort)f(D\))p Fl(.)e(This)i(coun)o(ter)h(is)f (incremen)o(ted)p 1650 550 V 83 w(+1)p 1882 550 V 1891 550 V 28 600 V 37 600 V 313 600 V 339 585 a(b)o(y)f(1)h(when)g(an)g(instruction)g (bu\013er)h(fetc)o(h)f(is)g(initiated.)p 1650 600 V 1882 600 V 1891 600 V 29 602 1863 2 v 28 652 2 50 v 37 652 V 62 637 a(3)p 313 652 V 256 w Fm(Floating-p)q(oi)o(n)n(t)f(add)i(op)q(erations)p Fl(.)g(This)f(coun)o(ter)h(is)f(incremen)o(ted)g(b)o(y)g(1)p 1650 652 V 86 w(+1)p 1882 652 V 1891 652 V 28 702 V 37 702 V 313 702 V 339 687 a(when)g(a)g(result)g(is)g(pro)q(duced)h(from)d(the)j (\015oating)e(p)q(oin)o(t)g(addition)g(functional)p 1650 702 V 1882 702 V 1891 702 V 28 751 V 37 751 V 313 751 V 339 736 a(unit.)k(This)d(includes)g(b)q(oth)h(scalar)e(and)h(v)o(ector)h(mo)q(de)e (execution.)p 1650 751 V 1882 751 V 1891 751 V 29 753 1863 2 v 28 803 2 50 v 37 803 V 62 788 a(4)p 313 803 V 256 w Fm(Floating-p)q(oi)o (n)n(t)g(m)o(ultip)o(ly)g(op)q(erations)p Fl(.)i(This)f(coun)o(ter)h(is)f (incremen)o(ted)p 1650 803 V 80 w(+1)p 1882 803 V 1891 803 V 28 853 V 37 853 V 313 853 V 339 838 a(b)o(y)f(1)h(when)g(a)g(result)g(is)g (pro)q(duced)h(from)d(the)j(\015oating)e(p)q(oin)o(t)g(m)o(ultiply)p 1650 853 V 1882 853 V 1891 853 V 28 902 V 37 902 V 313 902 V 339 888 a(functional)f(unit.)18 b(This)c(includes)g(b)q(oth)g(scalar)g(and) g(v)o(ector)h(mo)q(de)e(execution.)p 1650 902 V 1882 902 V 1891 902 V 29 904 1863 2 v 28 954 2 50 v 37 954 V 62 939 a(5)p 313 954 V 256 w Fm(Floating-p)q(oi)o(n)n(t)g(recipro)q(cal)h(op)q(erations)p Fl(.)h(This)f(coun)o(ter)h(is)f(incremen)o(ted)p 1650 954 V 50 w(+1)p 1882 954 V 1891 954 V 28 1004 V 37 1004 V 313 1004 V 339 989 a(b)o(y)f(1)h(when)g(a)g(result)g(is)g(pro)q(duced)h(from)d(the)j (\015oating)e(p)q(oin)o(t)g(recipro)q(cal)p 1650 1004 V 1882 1004 V 1891 1004 V 28 1054 V 37 1054 V 313 1054 V 339 1039 a(appro)o(ximation)d(functional)j(unit.)18 b(This)c(includes)g(b)q(oth)g (scalar)g(and)g(v)o(ector)p 1650 1054 V 1882 1054 V 1891 1054 V 28 1103 V 37 1103 V 313 1103 V 339 1088 a(mo)q(de)e(execution.)p 1650 1103 V 1882 1103 V 1891 1103 V 29 1105 1863 2 v 28 1155 2 50 v 37 1155 V 62 1140 a(6)p 313 1155 V 256 w Fm(CPU)j(memory)g(references) f(\(p)q(orts)h(A,)h(B,)g(C\))p Fl(.)d(This)h(coun)o(ter)h(is)f(incre-)p 1650 1155 V 71 w(+3)g(max)p 1882 1155 V 1891 1155 V 28 1205 V 37 1205 V 313 1205 V 339 1190 a(men)o(ted)f(b)o(y)h(the)g(sum)f(of)g(the)i (CPU)f(memory)d(references)17 b(from)12 b(p)q(orts)j(A,)e(B,)p 1650 1205 V 1882 1205 V 1891 1205 V 28 1254 V 37 1254 V 313 1254 V 339 1240 a(and)g(C.)h(This)f(includes)i(b)q(oth)f(scalar)g(and)f(v)o (ector)i(mo)q(de)e(execution.)p 1650 1254 V 1882 1254 V 1891 1254 V 29 1256 1863 2 v 28 1306 2 50 v 37 1306 V 62 1291 a(7)p 313 1306 V 256 w Fm(I/O)i(memory)g(references)f(\(p)q(ort)h(D\))p Fl(.)d(This)i(coun)o(ter)h(is)f(incremen)o(ted)p 1650 1306 V 123 w(+1)p 1882 1306 V 1891 1306 V 28 1356 V 37 1356 V 313 1356 V 339 1341 a(b)o(y)f(1)h(for)f(eac)o(h)i(I/O)f(memory)d(reference)16 b(accessed)h(through)c(this)h(CPU.)p 1650 1356 V 1882 1356 V 1891 1356 V 29 1357 1863 2 v 363 1432 a(T)m(able)f(1:)18 b(CRA)m(Y)13 b(Y-MP)h(Hardw)o(are)h(P)o(erformance)e(Monitor)h(Group)f(0.)-90 1599 y Fh(2.2)56 b(Hardw)n(are)19 b(P)n(erformance)e(Monitor)i(Recording)e (Metho)r(dology)-90 1725 y Fl(In)d(June)g(1991,)e(the)j(National)d(Cen)o(ter) j(for)e(Sup)q(ercomputing)g(Applications,)g(an)g(NSF-supp)q(orted)i(facilit)o (y)d(at)i(the)g(Univ)o(ersit)o(y)g(of)-90 1825 y(Illinois)d(at)h (Urbana-Champaign,)e(installed,)i(at)g(the)i(request)g(of)e(one)h(of)f(the)h (authors\(JL\),)g(a)g(mo)q(di\014cation)d(to)j(the)g(UNICOS)g(6.0)-90 1924 y(op)q(erating)h(system)f(on)h(their)g(CRA)m(Y)g(Y-MP/4.)j(This)d(c)o (hange)g(caused)h(UNICOS)g(to)f(write)g(HPM)g(data)g(to)f(a)h(system)g (\014le)g(at)f(the)-90 2024 y(termination)i(of)g(eac)o(h)i(user)g(pro)q(cess) h(or)f(job)e([15)o(].)25 b(Jobs)17 b(that)f(w)o(ere)h(not)g(link)o(ed)e (after)i(June)g(1,)f(1991)f(are)i(not)f(recorded.)27 b(This)-90 2123 y(includes)12 b(some)f(third)h(part)o(y)f(soft)o(w)o(are)h(whic)o(h)f(w) o(as)h(installed)f(in)g(a)g(binary)h(form.)j(A)d(\014lter)g(w)o(as)f(emplo)o (y)o(ed)f(to)i(a)o(v)o(oid)e(the)i(recording)-90 2223 y(of)17 b(small)d(pro)q(cesses)20 b(taking)c(less)i(than)f(one)h(second,)g(e.g.)28 b(user)18 b(commands)d(suc)o(h)j(as)f('ls'.)27 b(The)17 b(statistics)h(accum) o(ulated)f(in)f(a)-90 2323 y(system)e(\014le)f(that)h(ma)o(y)e(b)q(e)i(p)q (ost-pro)q(cessed,)i(summarized,)c(and)h(sorted)i(in)e(v)n(arious)g(w)o(a)o (ys)h(to)f(pro)q(duce)i(rep)q(orts)h(ab)q(out)d(activities)-90 2422 y(that)f(o)q(ccurred)h(during)f(the)g(recording)g(p)q(erio)q(d)g([13)o (].)17 b(The)12 b(UNICOS)h(HPM)f(collection)f(co)q(de)i(on)e(the)h(Y-MP)g (gathers)h(information)-90 2522 y(organized)h(in)f(three)j(records,)f(called) e(A,)h(B)g(and)g(C,)f(ab)q(out)h(eac)o(h)g(user)h(job:)76 2604 y Fm(A)f Fl(hpmglobal)d Fm(timestamp)i(uid)h(pid)h(acct)76 2654 y(B)h(cmdname)76 2704 y(C)g(group)f(c0)h(c1)g(c2)g(c3)g(c4)g(c5)g(c6)g (c7)g(total)950 2828 y Fl(6)p eop %%Page: 7 9 bop -90 195 a Fl(where)76 278 y Fm(timestamp)11 b Fl(is)j(the)h(date/time)d (stamp,)h(as)g(pro)o(vided)h(b)o(y)g(\\time\(\)".)76 328 y Fm(uid)e Fl(is)i(the)g(user's)h(ID)f(n)o(um)o(b)q(er,)f(as)h(pro)o(vided)f(b) o(y)h(\\getuid\(\)".)76 378 y Fm(pid)e Fl(is)i(the)g(pro)q(cess)i(n)o(um)o(b) q(er,)d(as)h(pr