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

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