CHARACTERIZING DISTRIBUTED SHARED-MEMORY PERFORMANCE - A CASE-STUDY OF THE CONVEX SPP1000

In a distributed shared memory (DSM) multiprocessor, the processors co operate in solving a parallel application by accessing the shared memo ry. The latency of a memory access depends on several factors, includi ng the distance to the nearest valid data copy, data sharing condition s, and traffic of other processors. To provide a better understanding of DSM performance and to support application tuning and compiler deve lopment for DSM systems, this paper extends microbenchmarking techniqu es to characterize the important aspects of a DSM system. We present a n experiment-based methodology for characterizing the memory, communic ation, scheduling, and synchronization performance, and apply it to th e Convex SPP1000. We present carefully designed microbenchmarks to cha racterize the performance of the local and remote memory, producer-con sumer communication involving two or more processors, and the effects on performance when multiple processors contend for utilization of the distributed memory and the interconnection network.