A Parallel Recursive H/V Partitioning Method for Mapping Unstructured Finite Element Graphs on Processor Meshes

Abstract

To efficiently execute a finite element modeling program on a distributed memory multicomputer, we need to distribute the takes of a finite element graph tp the processors of a distributed memory multicomputer as evenly as possible and minimize the communication cost of processors. In this paper, we present a parallel recursive H/V (Horizontal/Vertical) partitioning method to efficiently map unstructured finite element graphs on processor meshes. Given an unstructured finite element graph and an m x n processor mesh, this method tries to balance the computation load and minimize the communication cost simultaneously by recursively applying the horizontal/vertical partitioning method to divide the unstructured finite element graph and the processor mesh into two subgraphs and two sub-processor meshes, respectively, until all sub-processor meshes contain one processor. To evaluate the performance of the parallel recursive H/V partitioning method, we have implemented the proposed method on simulated processor meshes alone with the nearest-neighbor mapping method[17]. Five unstructured finite element graphs are used as test samples. The simulation results show that the proposed method outperforms the nearest-nrighbor mapping method for all test samples.