Performance Evaluation of Message Passing vs. Multi-threading Parallel Programming Paradigms on Multi-core Systems

Abstract

Present and future multi-core computational system architecture attracts researchers as an adequate and inexpensive solution to achieve high performance computation for many problems. The multi-core architecture enables implementation of shared memory and/or message passing parallel processing paradigms. Therefore, there is a great need for standard libraries in order to utilize the resources efficiently and effectively. In this work, we evaluate the performance of message passing using two versions of the well-known message-passing interface (MPI) library: MPICH1 vs. MPICH2. Furthermore, we compared the performance of shared memory using OpenMP that supports multi-threading with MPI. The added features (total 9) impacted the MPICH2 results over MPICH1. On the other hand, the overheads of message passing and large data communication impact negatively on the performance of this paradigm against multi-threading paradigm.