当前位置:主页 > 科技论文 > 计算机论文 >

高阶矩量法在国产超级计算机上的并行性能

发布时间:2019-05-12 19:33
【摘要】:矩量法的计算复杂度和内存需求随着电磁目标未知量N的增大呈O(N 2)甚至O(N 3)增长,这使得其在处理未知量很大的复杂电磁问题时,往往需要很长的计算时间,因此算法并行性能的优化显得尤为重要。本文基于国产计算机平台的软硬件环境,针对Block size和Process grid两个参数对矩量法并行性能的影响进行了优化测试,并根据优化测试的结论对高阶矩量法的并行性能进行了测试,测试结果表明该算法在并行规模扩大10倍以上时仍可以获得75%~80%的并行效率。此外以飞机散射特性计算为例表明并行高阶矩量法可以在国产计算机平台上有效解决此类散射问题。
[Abstract]:The computational complexity and memory requirement of the moment method increase with the increase of the unknown quantity N of the electromagnetic target, which is O (N 2) or even O (N 3), which makes it often take a long time to deal with the complex electromagnetic problems with large unknowns. Therefore, the optimization of parallel performance of the algorithm is particularly important. Based on the software and hardware environment of the domestic computer platform, this paper tests the parallel performance of the method of moment (mom) under the influence of Block size and Process grid parameters, and tests the parallel performance of the high-order mom according to the conclusion of the optimization test. The test results show that the parallel efficiency of the algorithm can still be 75% 鈮,

本文编号:2475644

资料下载
论文发表

本文链接:https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2475644.html


Copyright(c)文论论文网All Rights Reserved | 网站地图 |

版权申明:资料由用户8b4f5***提供,本站仅收录摘要或目录,作者需要删除请E-mail邮箱bigeng88@qq.com