Evaluting Performance of OpenMP and MPI on the SGI Origin 2000 with Benchmarks of Realistic Problem Sizes

Abstract

Six applications benchmarks, including four NAS codes, provided by H. Jin and J. Wu, previously parallelized using OpenMP, and message-passing-interface (MPI), were run on a 128-processor SGI Origin 2000. Detailed profile data was collected to understand the factors causing imperfect scalability. Our results show that load imbalance and cost of remote accesses are the main factors in limited speed-up of the OpenMP versions, whereas communication costs are the single major factor in the performance of the MPI versions.