摘要
针对采用有限元方法计算复杂、电大尺寸目标时,面临的内存不足、计算耗时过长的问题,研究了有限元方法的大规模并行化实现。通过使用图分割理论的均衡网格划分策略以及稀疏矩阵重排序技术来提高程序的性能和可扩展性。在我国天河二号超级计算机上成功突破1000 CPU核并行有限元计算,并行效率达到65%以上。
Focusing on the problems of insufficient memory and too long consuming time when using finite element method to calculate complex structure and electrically large targets, the implementation of large-scale finite element method is studied in this paper. The balanced meshing strategies that based on graph segmentation theory, and the sparse matrix reordering techniques are adopted to improve the performance and scalability of the program. The parallel scale of finite element method is successfully exceeded to 1,000 CPU cores on the Tianhe-2 supercomputer, and the parallel efficiency is above 65%.
引文
[1]JinJ.TheFiniteElementMethodin Electromagnetics[M]. New Jersey:JOHN WILEY&SONS, 2002.
[2]张玉,赵勋旺,陈岩等.计算电磁学中的超大规模并行矩量法[M].西安电子科技大学出版社, 2016.
[3]L. E. Garcia-Castillo, I. Gómez-Revuelto, F. Sáez de Adana, and M. Salazar-Palma,“A finite element method for the analysis of radiation and scattering of electromagnetic waves on complex environments,”ComputerMethodsinApplied Mechanics and Engineering, vol. 194/2-5, pp. 637–655, Feb. 2005.
[4]Garcia-Donoro D, Garcia-Castillo L E, Ting S W.Verification Process of Finite-Element Method Code for Electromagnetics:Using the method of manufacturedsolutions[J].IEEEAntennas&Propagation Magazine, 2016, 58(2):28-38.