nPME对GROMACS软件并行计算性能的影响分析

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。%Correctly dealing with the task allocation of both parallel scale and PME method in parallel molecular dynamics computation has critical influence on raising the parallel efficiency of molecular dynamics.We take the commonly used molecular dynamics software“Gromacs”for example,and make use the“MagicCube”supercomputer platform in Shanghai Supercomputer Center,carry out the tests on the computation performances under different parallel scales and different PME task allocation modes.It is found that the parallel can significantly raise compu-tation efficiency of Gromacs within a certain range.However,when the parallel scale exceeds a certain value,the computation efficiency of Gro-macs will decrease instead.On the other hand,when the parallel scale is small,the computation efficiency without setting the exclusive PME nodes will be higher;while in larger parallel scale,it is able to remarkably raise the computation efficiency by reasonably setting the exclusive nodes of PME.This result can provide helps for the researchers engaging in molecular dynamics simulation in reasonable allocating the parallel resources.