国产CPU平台中高阶矩量法10万核并行性能
详细信息    查看全文 | 推荐本文 |
  • 英文篇名:The Hundred Thousand Cores Parallel Performance of High Order Mo M in Domestically-Made CPU Platform
  • 作者:左胜 ; 林中朝 ; 陈岩 ; 张玉 ; 赵勋旺
  • 英文作者:ZUO Sheng;LIN Zhong-chao;CHEN Yan;ZHANG Yu;ZHAO Xun-wang;Key Lab. of Antennas and Microwave Technology,Xidian Univ.;
  • 关键词:矩量法 ; 高阶基函数 ; 并行效率 ; 超级计算机 ; 10万核
  • 英文关键词:method of moments;;high order basic function;;parallel efficiency;;super computer;;100k CPU cores
  • 中文刊名:WBXB
  • 英文刊名:Journal of Microwaves
  • 机构:西安电子科技大学天线与微波技术重点实验室;
  • 出版日期:2017-12-21 14:34
  • 出版单位:微波学报
  • 年:2017
  • 期:v.33
  • 基金:国家重点研发计划政府间国际科技创新合作重点专项(2016YFE0121600);; 教育部新世纪优秀人才支持计划(NCET-13-0949);; 中央高校基本科研业务费(JB160218);; 国家高技术研究发展计划(863计划)(2012AA01A308)
  • 语种:中文;
  • 页:WBXB201706018
  • 页数:4
  • CN:06
  • ISSN:32-1493/TN
  • 分类号:85-88
摘要
针对国产超级计算机平台上大规模电磁仿真软件相对匮乏,本文将并行高阶矩量法程序移植到国产超级计算机平台上,并以机载线天线阵列的辐射特性计算为例对其并行性能进行了测试和评估。实现了并行高阶矩量法单一任务突破10万CPU核规模,这是目前在国产超级计算机平台上实现的最大规模并行矩量法计算。以1440核为基准,使用CPU核数达到102400,并行规模扩大约70倍时,并行矩量法矩阵方程求解并行效率仍在50%以上。这一研究工作,使利用纯国产超级计算机对复杂电大电磁系统进行精确高效仿真成为可能。
        To solve the deprivation of large-scale electromagnetic simulation software in domestically-made CPU platform,the paper translates the parallel higher-order Mo M program into domestically-made CPU platform and program's parallel performance is tested and evaluated by the calculation of the radiation characteristic of the airborne wire antenna array. The maximum number of CPU cores breaks through 100 k cores,which is the largest scale of parallel Mo M computation in domestically-made CPU platform at present. The results show that the parallel efficiency is higher than 50% when the parallel scale is expanded about 70 times from 1440 CPU cores to 102400 CPU cores. This study makes it possible that make use of the pure domestic super computer to complete the accurate and efficient simulation of complex electrically large electromagnetic system.
引文
[1]Zhang Y,Tapan K Sarkar.Parallel Solution of Integral Equation-Based EM Problems in the Frequency Domain[M].Hoboken,NJ:Wiley-IEEE Press,2009
    [2]Zhang Y,Van d G R A,Taylor M C,et al.Parallel Mo M using higher-order basis functions and PLAPACKin-core and out-of-core solvers for challenging EM simulations[J].IEEE Antennas&Propagation Magazine,2009,51(5):42-60
    [3]Lin Z,Zhang Y,Jiang S,et al.Simulation of airborne antenna array layout problems using parallel higher-order Mo M[J].International Journal of Antennas&Propagation,2014,2014(9):1-11
    [4]Zhang Y,Lin Z,Zhao X,et al.Performance of a massively parallel higher-order method of moments code using thousands of CPUs and its applications[J].IEEE Transactions on Antennas&Propagation,2014,62(12):6317-6324
    [5]林中朝,张爽,王星,等.高阶矩量法在国产超级计算机上的并行性能[A].全国军事微波技术暨太赫兹技术学术会议[C],2014Lin Z,Zhang S,et al.The parallel performance of higher-order Mo M on a domestically-made CPU platform[A].National Conference on Microwave and Terahertz Technology[C],2014
    [6]林中朝,陈岩,张玉,等.国产CPU平台中并行高阶矩量法研究[J].西安电子科技大学学报,2015(3):43-47Lin Z,Chen Y,Zhang Y,et al.Study of the parallel higher-order Mo M on a domestically-made CPU platform[J].Journal of Xidian University,2015,42(3):43-47
    [7]张玉.计算电磁学中的超大规模并行矩量法[M].西安:西安电子科技大学出版社,2016Zhang Y.Oversize Parallel Mo M in Computational Electromagnetics[M].Xi'an:Press of Xi'an University of Electronic Science and Technology,2016
    [8]Netlib Repository at UTK and ORNL,The Scal APACKProject,Parallel Basic Linear Algebra Subprograms(PBLAS)Home Page.Available at:http://www.netlib.org/scalapack/pblas_qref.html