Parallel FMM Algorithm Based on Space Decomposition

In this paper, a parallel computational model and algorithm based on space decomposition is constructed and implemented, which supports the dynamically resource allocation under cluster environment. The major aim is to explore the new space decomposition scheme that can solve computation intensive problem. The fast multipole method (FMM) is an algorithm for rapid evaluation of the potential and force fields in the system involving large numbers of particles. Based on the serial FMM algorithm, a parallel implementation entitled SDPFMM is presented in the paper. The proposed algorithm is characterized by scalability and flexibility. We carried out the experiment on the high-performance computer ZQ3000 with Intel Trace Analyzer and Collector integrated into SDPFMM, and analyzed the experimental data and MPI performance of SDPFMM. The results demonstrate that the proposed algorithm is satisfying in both efficiency and solution quality.