nPME对GROMACS软件并行计算性能的影响分析

发布时间：2018-05-01 23:04

本文选题：并行计算 + 分子动力学　；参考：《计算机应用与软件》2014年10期

【摘要】：在分子动力学并行计算的过程中,正确地处理好并行规模与PME(Particle-Mesh Ewald)方法的任务分配,对于提高分子动力学的并行效率具有非常重要的影响。以常用的分子动力学软件Gromacs[1-3]为例,利用上海超级计算中心"魔方"超级计算平台,就不同并行规模与不同PME任务分配方式下的计算性能进行测试。发现并行能在一定的范围内显著的提高Gromacs的计算效率,但是当并行规模超过一定值后,计算效率反而下降。另一方面,当并行规模较小时,不设置PME专属节点的计算效率较高;但当并行规模较大时,合理的设置PME专属节点,能显著地提高计算的效率。这一结果能对从事分子动力学模拟的研究人员合理分配并行资源提供帮助。
[Abstract]:In the process of the parallel computation of molecular dynamics, the correct handling of the task allocation of parallel scale and PME (Particle-Mesh Ewald) method has a very important influence on improving the parallel efficiency of molecular dynamics. The common molecular dynamics software Gromacs[1-3] is used as an example to make use of the "magic square" supercomputing of the Shanghai Supercomputing Center. Test performance is tested on different parallel scale and different PME task allocation methods. It is found that parallel can significantly improve the computing efficiency of Gromacs in a certain range, but when the parallel scale exceeds a certain value, the computational efficiency decreases. On the other hand, when the parallel scale is small, the computing efficiency of the PME exclusive node is not set. The rate is high, but when the parallel size is large, the reasonable setting of the PME exclusive node can significantly improve the efficiency of the calculation. This result can help the researchers in the molecular dynamics simulation to allocate the parallel resources reasonably.

【作者单位】：上海超级计算中心;复旦大学物理系理论生命研究中心;
【基金】：国家高技术研究发展计划项目(2012AA01A308) 上海市科委项目(1201H116400)
【分类号】：TP338.6

【参考文献】