Performance Comparison Of Parallel Algorithms On Different Communication Architectures

The objective of this work is to compare the performance of three common environments for supporting parallel computing, by focusing on the impact of communications on this performance. In particular, the studied environments include a shared memory multiprocessor, a distributed memory multiprocessor, and an ATM-based network of workstations. The measurements consist of the execution of real parallel codes, obtained from well-known benchmark suites. Our results show that the characteristics of communications in each parallel code significantly determine its best-suited parallel computing environment. Competitive performance can be achieved on the ATM-based environment provided that some bottlenecks in the network and the hosts are reduced their impact. Keywords: Communications in Parallel Computing,Shared MemoryMultiprocessors, Distributed Memory Multiprocessors, Networks of Workstations, ATM (Asynchronous Transfer Mode) 1 INTRODUCTION The advent of multiprocessor systems has enabled..