next up previous contents
Next: 3 Cluster computing assumptions Up: No Title Previous: 1 Introduction

2 Cluster Efficiency

 

The C90 computer itself is already a parallel computer (with shared memory). In order to measure the effects of clustering several C90 systems rather than measuring the overall parallel efficiency (counting all computing nodes involved), we define the following expression as cluster speed-up

eqnarray51

and as cluster efficiency

eqnarray57

Here P denotes the number of supercomputers (C90 systems) used in the cluster and tex2html_wrap_inline1868 the computing time of the parallel code on a cluster consisting of P systems.

For the parallel IFS code, we consider the 2D partitioning consisting of tex2html_wrap_inline1870 processes. We assume that the parallel code is partitioned to the cluster situation in an appropriate way. tex2html_wrap_inline1868 represents always the shortest runtime among all partitionings and mappings implemented with the IFS version RAPS 2.0.



top500@rz.uni-mannheim.de
Tue May 28 14:38:25 PST 1996