An improved partitioning strategy for structured multiblock grids

Abstract

Load balancing is one of the critical factors affecting the performance of parallel computing. An improved partitioning strategy is proposed for structured multiblock grids here. The new subgrid-splitting approach, together with a recursive graph-partitioning algorithm, is to seek a good load balancing as less blocks as possible. Two typical applications are implemented to validate the strategy. Results demonstrate that the new partitioning strategy behaves well in load balancing and communication overheads, and furthermore manifests an excellent performance in decreasing the amount of blocks, as well as the memory requirement caused by the multiplication of ghost cells near the edge.