• 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 •    下一篇

基于Intel MIC平台大规模耗散粒子动力学模拟的设计与优化

徐顺1,2,刘倩1,2,张宝花1,2,何连花1,2,金钟1,2   

  1. (1.中国科学院计算机网络信息中心,北京 100190;2.中国科学院计算科学应用研究中心,北京 100190)
  • 收稿日期:2017-01-13 修回日期:2017-04-27 出版日期:2017-08-25 发布日期:2017-08-25
  • 基金资助:

    国家重点研发计划“高性能计算”重点专项项目(2016YFB0201700);中国科学院-英特尔并行计算中心项目(902016000200);中国科学院青年创新促进会(2016156)

Design and optimization of large scale dissipative particle dynamics simulation on Intel MIC platform

XU Shun1,2,LIU Qian1,2,ZHANG Bao-hua1,2,HE Lian-hua1,2,JIN Zhong1,2   

  1. (1.Computer Network Information Center,Chinese Academy of Sciences,Beijing 100190;
    2.Center of Scientific Computing Applications & Research,Chinese Academy of Sciences,Beijing 100190)
     
  • Received:2017-01-13 Revised:2017-04-27 Online:2017-08-25 Published:2017-08-25

摘要:

耗散粒子动力学(DPD)模拟是一种重要的研究流体动力学特性的计算模拟方法,基于Intel MIC平台设计实现了面向大规模耗散粒子动力学模拟,充分结合了DPD模拟本身的特性和MIC平台的特征。对DPD模拟中的近邻列表构建和短程作用力关键代码实现了向量化优化,在CPU和MIC协处理器之间采用任务计算负载平衡机制,支持MPI进程内线程数量负载平衡控制。分别在原型程序上和LAMMPS集成中做了性能对比分析,实验结果显示了引入相关优化技术的有效性,为进一步研究面向MIC众核平台的分子动力学相关工作奠定了基础。
 
 

关键词: 耗散粒子动力学, 众核, 图像处理单元, 分子动力学

Abstract:

Dissipative particle dynamics (DPD) simulation is an important computational simulation method to investigate the dynamic characteristics of the fluid. We design and carry out an Intel MIC platform-based large-scale dissipative particle dynamics simulation. The design combines the characteristics of DPD simulation and the features of MIC platform. The critical codes of DPD simulation such as neighbor list building and calculation of short-range force are designed with the vector optimization. A load balancing mechanism of tasks between the CPU and MIC co-processors is applied to support the load balancing control of the number of threads within a MPI process. The performance analysis on a prototype program and LAMMPS integration program is compared, and the experimental data show the effectiveness of introducing relevant optimization technologies. This work lays the foundation for further research on molecular dynamics simulations on MIC many-core platform.

Key words: dissipative particle dynamics, many-core, GPU, molecular dynamics