OpenMP vs. MPI on a Shared Memory Multiprocessor

摘要

Porting a parallel application from a distributed memory system to a shared memory multiprocessor can be done by reusing the existing MPI code or by parallelising the serial version with OpenMP directives. These two routes are compared for the case of the climate model ECHAM5 on a IBM pSeries690 system. It is shown, that in both cases modifications of computation and communication patterns are needed for high parallelisation efficiency. The best OpenMP version is superior to the best MPI version, and has the further advantage of allowing a very efficient loadbalancing.