Optimizing performance of parallel programs on multicomputer and multi-core architectures

Abstract

With the advent of multi-core architectures, there arises a need for comparative evaluations of the performance of well-understood parallel programs. This is because, it is necessary to gain an insight into the potential advantages of the available platforms, namely multi-core and multi-processors, to decide which platform to use for a particular application. The need for this insight is due to the different nature and requirements (like the divisions of work, intercommunications etc.) of the different parallel algorithms. In this paper, we evaluate the performance of the parallel implementations of three well known algorithms on multi-computer, multi-core and hyper-threading architectures. Parallelization of the programs was done using MPICH2 and OpenMP. We provide comparative evaluation of the run-time behaviors of the parallel programs using three performance metrics: average run-time, I/O and communication overhead. The main experimental result demonstrates the superiority of multicore architecture over multi-computer and hyper-threading architectures for running threaded applications for the same number of cores/processors. We also investigate the effect of parallel I/O in the performance of the programs in multi-computer platform.