大规模并行RWG矩量法矩阵填充优化

Optimization of matrix filling in the large scale parallel RWG basis based method of moments

针对并行RWG矩量法进程间冗余积分问题,通过优化网格编号提出了一种高效的并行矩阵填充方案.在矩阵块循环分布并行策略基础上,对三角形公共边进行重新编号,使得需要相同三角形积分的矩阵元素分布在同一进程上,从而大幅度地减少进程间的冗余积分计算.数值结果表明,该并行矩阵填充方案消除了绝大部分的进程间冗余积分,提高了并行矩阵填充的效率.

To solve the issue of inter-process redundant integrals in the parallel Method of Moments （MoM） using RWG basis functions, an efficient parallel matrix filling scheme is proposed through mesh index optimization. Based on a block-cyclic matrix distribution strategy, the common edges of triangles are renumbered to make the matrix elements that need the same triangular integrals be assigned to one process, thus drastically reducing the inter-process redundant integrals. Numerical results show that the proposed scheme eliminates most of the inter-process redundant integrals and greatly improves the efficiency of parallel matrix filling.