• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于二维结构化网格的可压缩流体并行算法研究

皇甫永硕,刘杰,龚春叶   

  1. (国防科技大学并行与分布处理国家重点实验室,湖南 长沙 410073)
  • 收稿日期:2015-12-01 修回日期:2016-01-22 出版日期:2017-09-25 发布日期:2017-09-25
  • 基金资助:

    国家自然科学基金(61402039,91430218);国家863计划(2012AA01A301);博士后基金(2014M562570,2015T81127)

A parallel compressible fluid algorithm based on
two-dimensional structural grid

HUANGFU Yong-shuo,LIU Jie,GONG Chun-ye   

  1. (National Laboratory of Parallel and Distributed Processing,
    National University of Defense Technology,Changsha 410073,China)
  • Received:2015-12-01 Revised:2016-01-22 Online:2017-09-25 Published:2017-09-25

摘要:

基于二维/轴对称高精度可压缩多相流计算流体力学方法MuSiC-CCASSIM的结构化网格部分,设计了区域并行分解方法;针对各处理器边界数据的通信,设计了阻塞式通信与非阻塞式通信并行算法;为了减少通信开销,设计了MPI/OpenMP混合并行优化算法。在天河二号超级计算机上进行了测试,每个核固定网格规模为625*250,最多调用8 192核。测试数据表明,采用MPI/OpenMP混合并行算法、纯MPI非阻塞式通信并行算法和纯MPI阻塞式通信并行算法的程序的平均并行效率分别达到86%、83%和77%,三种算法都具有良好的可扩展性。
 

关键词: 并行计算, 阻塞式通信, 非阻塞式通信, MPI/OpenMP混合编程, 天河计算机

Abstract:

Based on the two-dimensional/axisymmetric high-precision compressible multiphase flow computational fluid dynamics method  multi-scale simulation code  (MuSiC)-consistent, conservative, all-speed, sharp-interface method (CCASSIM), we propose a parallel area decomposition  method to deal with structural grids. According to different communication ways of boundary data on each processor, blocking and non-blocking communication algorithms are designed respectively. We also develop a MPI / OpenMP hybrid parallel optimization algorithm to reduce communication overhead. Some  numerical experiments are performed on Tianhe-2 supercomputer system with the maximum of grid cells of 1.2 × 1010 and 8 192 processors. The results show that the average parallel efficiency of the program that uses the MPI / OpenMP hybrid parallel algorithm, the pure MPI parallel algorithm with non-blocking communication, and the pure MPI parallel algorithm with blocking communication is 86%, 83% and 77% respectively. Futhermore, good scalability is also observed from the test results of the above three algorithms.
 

Key words: parallel computing, blocking communication, non-blocking communication, MPI/OpenMP hybrid programming, Tianhe supercomputer