Comparison of Parallelization Models for Structured Adaptive Mesh Refinement

摘要

We compare different parallelization approaches in different programming models using MPI and OpenMP on a SunFire 15K for structured adaptive mesh refinement. The results show that no static parallelization or programming model performs best in all cases and that different models are preferable in different settings of the application and grid hierarchy. We then suggest a dynamic approach using a mixed MPI-OpenMP model that adapts the number of threads during run time and gives good performance in all stages throughout the whole run as the solution state changes, i.e. the resolution in the computational grid changes.