Portable Parallel Simulator using MPI for 2D and 3D Domains: Design and Performance Testing

Abstract. In this paper, we present the design and implementation details of our highly modular parallel simulation code using MPI, overlapping communication and computations. We outline the modularity of the implementation, potentially allowing the code to be adapted for other applications. For the testbed Itanium2-based 256-processor running Linux cluster at the Academic Computer Centre TASK, Gdansk, Poland, we present the relation between the speed-ups and do-main sizes as well as shapes of 3D domains with different ratios of nodes up-dated by each processor to the number of nodes exchanged with the processors holding neighboring domains. We also present the relation between the parallel efficiency and the idle time spent on finishing pending communication requests, for all tested domain configurations, from 1 to 32 processors. The latter could be beneficial for advance assessment of other applications run on the same cluster.