Electromagnetic Problem Based On A Distributed Network Of Parallel Computing Research

The distributed network parallel computation is one of the key development directions in high performance computation. In this dissertation, the parallel algorithms and their application in the electromagnetic field are discussed in detail. Based on MPI in the distributed-memory cluster system, CG and CG-ADDM parallel algorithms are presented to solve large sparse linear systems of finite element method (FEM) derived from waveguide and microstrip models.Conjugate Gradient (CG) algorithm plays an important role in solving large sparse linear systems. The processing of matrix-vector multiplication is designed to be parallel for the same effectiveness of CG algorithm. The quantity of communication data among processes is also considered. CG algorithm is optimized by using adaptive communication algorithm and non-blocking communication method. The simulation results of waveguide model demonstrate that its parallel performance is excellent.CG-ADDM algorithm is the effective combination of CG algorithm and ADDM(Algebraic Domain Decomposition Method). To solve large sparse linear system, the proposed algorithm segments the problem into several smaller sub-problems with different granularity. The whole results are obtained by iterative parallel steps of solving each sub-problem in each node (i.e. computer) by direct method, exchanging related data among nodes in MPI cluster network and reassembling the sub-problem solutions together to obtain the global results of the reduced system. The simulation of large sparse linear system based on waveguide and microstrip models achieves excellent parallel performance.