基于深腾1800机群系统的分子动力学并行仿真研究  

Study of MD parallel simulation based on Shenteng 1800COW

作　　者：郭晓光[1] 郭东明[1] 康仁科[1] 金洙吉[1] 

机构地区：[1]大连理工大学精密与特种加工教育部重点实验室,辽宁大连116024

出　　处：《大连理工大学学报》2007年第6期818-822,共5页Journal of Dalian University of Technology

基　　金：国家自然科学基金资助项目(重大项目50390061);国家杰出青年科学基金资助项目(50325518)

摘　　要：介绍了分子动力学并行仿真计算的软硬件环境,分析了现有的几种并行算法,确定采用区域分解法作为并行算法,并在此基础上提出了基于区域二次划分的分子动力学并行仿真算法.另外,阐述了原子链、原子近邻表和原子亲属表的概念,提出了基于永久序号的消息传递策略.最后,设计了分子动力学并行仿真程序,并分别在1、2、3、4台结点机上进行了实验,运行结果表明:加速比随着结点数的增加而增加,并行效率虽略有下降但都在87.5%以上,并行效率并没有随着结点数的增加有明显的降低,说明并行程序具有很好的扩展性.The hardware and software environments for parallel algorithm of molecular dynamics （MD） simulation were discussed in detail firstly. Then some popular parallel algorithms were analyzed and the domain decomposition was selected for the parallel algorithm of MD simulation. Based on the domain decomposition, a twice domain partition algorithm was proposed for MD parallel simulation. Moreover, an atom chain, an atom neighbor list and an atom relative list were introduced, and an information visiting strategy based on the proposed permanent sequential number for each atom was designed. Lastly, MD parallel program was made and executed on 1,2,3,4 node computers respectively. The experimental results show that when the number of process increases, the accelerating ratio increases and the efficiency decreases a little but is still more than 87.5%. It means that the efficiency doesn＇t decrease obviously with the increase of the node number and the parallel program has good expansibility.

关 键 词：分子动力学仿真 并行算法 区域分解法 

分 类 号：TG580.1[金属学及工艺—金属切削加工及机床] TP311.1[自动化与计算机技术—计算机软件与理论]