SCIPVM: Parallel distributed computing on SCI workstation clusters

Workstation and PC clusters interconnected by SCT (scalable coherent interf ace) are very promising technologies for high-performance cluster computing . Using commercial SBus to SCI interface cards and system software and driv ers, a two-workstation cluster has been constructed for initial testing and evaluation. The PVM system has been adapted to operate on this cluster usi ng both raw channel and shared-memory access to the SCI interconnect, and p reliminary communications performance tests have been carried out, To achie ve mutual exclusion in accessing shared-memory segments, two protocols were used. Our preliminary results indicate that communications throughput in the rang e of 17.7 Mbytes/s, and round-trip latencies of 80 mu s using the first and 140 mu s using the second protocol, can be obtained on SCI clusters, These figures are significantly better (by a factor of 2 to 4) for small and lar ge messages than those attainable on Fast Ethernet LANs, Since these perfor mance results are very encouraging, we expect that, in the very near future , SCI networks will be capable of delivering several tens of Mbytes/s bandw idth and a few tens of microseconds latencies, and will significantly enhan ce the viability of cluster computing. Copyright (C) 1999 John Wiley & Sons , Ltd.