NA Digest Monday, May 18, 1987 Volume 87 : Issue 43

This weeks Editor: Gene Golub

Today's Topics:


Mail-From: GOLUB created at 17-May-87 22:40:43
Date: Thu, 14 May 87 09:34:22 edt
From: (Dennis Stevenson)
ReSent-Date: Sun 17 May 87 22:40:43-PDT
ReSent-From: Gene H. Golub <GGOLUB@Score.Stanford.EDU>
ReSent-To: na@Score.Stanford.EDU

Subject: USENET hypercube news group.

I am trying to set up a news group on USENET with the following outline:

Purpose. The purpose of the news group is to
(1) Provide a technical forum for exchange of information concerning
hypercubes. This information may be vendor specific or
vendor independent (such as general algorithms). Special
attention to scientific programming (not hardware oriented!).
(2) Provide a means for users to get to know one another.
(3) Provide for timely dissemination of special meetings and/or
activites directily relating to hypercubes.
(4) "LIMITED" announcements by manufacturers.

Focus. The focus will be on scientific programming on all manner of
hypercubes. Eventually, I would expect the focus to expand to
scientific programming issues on general non von Neumann architectures.

If you would like to see this news group come about, please respond to


Steve Stevenson
(803) 656-5880


Date: Mon, 18 May 87 10:15:01 EDT
From: mth@ORNL-MSR.ARPA (Michael T. Heath)
Subject: Parallel Short Course at ORNL

ORNL Short Course on Parallel Computing

The Mathematical Sciences Section of Oak Ridge National Laboratory will
offer a one-week short course on the practical use of multiprocessors
for solving computational problems in science and engineering. The
course is intended primarily for new staff members of ORNL and for
visiting faculty and students under various summer programs and the
upcoming Special Year in Numerical Linear Algebra, but others are also
welcome to attend subject to availability of space; priority will be
given to DOE-supported researchers. Primary emphasis will be on
distributed-memory multiprocessors, typified by hypercubes.
Participants will learn "hands-on" parallel programming techniques
using the Section's Intel iPSC d6 Hypercube, which has 64 processors
and 32 Mbytes of local memory. An NCUBE hypercube is also available
at ORNL, although it will not be the primary vehicle for the course.
In addition, parallel programming techniques for shared-memory
multiprocessors will be covered, using the Section's Sequent Balance
8000, which has 8 processors and 8 Mbytes of shared memory. All of
these machines have Unix-based operating systems and support both the C
and Fortran programming languages; prior familiarity with Unix would be
an extremely helpful but not absolutely necessary prerequisite.

Participants will be encouraged to gain practical experience by
completing a small project on each type of parallel architecture as
part of the course. In addition to a general orientation on both
machines, the course will also discuss important issues in parallel
algorithm development, including techniques for partitioning algorithms
for parallel implementation and efficient communication procedures.
Following the course, continued access to these parallel machines will
be provided to the extent that machine time is available. In order
that we may allocate these limited resources fairly, prospective users
may apply for access to either of the parallel machines by submitting a
brief proposal outlining a proposed project for parallel implementation.
Depending on demand, we anticipate offering the course twice, once in
summer and again in the fall. Tentative dates are June 22-26 and
September 28-October 2, 1987. For further information, please contact:

Mike Heath
Mathematical Sciences Section
Oak Ridge National Laboratory
P.O. Box Y, Bldg. 9207A
Oak Ridge, TN 37831
Phone 615-574-3130


End of NA Digest