近程作用分子动力学模拟的两级并行  

Two-level parallel computation of short-range molecular dynamics

作　　者：王小伟[1] 郭力[1] 杨章远[1] 

机构地区：[1]中国科学院过程工程研究所多相反应开放实验室,北京100080

出　　处：《计算机与应用化学》2003年第5期639-642,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金(20221603)

摘　　要：分子动力学作为一种重要的计算手段在许多领域有着广泛的应用,由于它的计算量比较庞大,因此并行计算方法被越来越多地引入到分子动力学的模拟中。本文在目前常见的SMP集群系统上,根据系统的结构特点,针对分子动力学的三种并行算法:区域分解法、原子分解法和力分解法,利用MPI+Pthread的混合编程模型,采用节点间消息传递模式以及节点内部共享存储的编程模式,实现了近程作用分子动力学的两级并行计算。计算结果表明,不同的算法采用了两级并行的方式和原来只有消息传递的并行方式相比,具有不同的计算效率,但是从总体来说采用两级并行的计算方式可以利用更多的计算资源,从而有助于提高计算能力。Classical molecular dynamics (MD) is an important computational tool widely used in many research fields. Parallel computation is more and more applied to MD simulation because of the giant computation in it. The mixed model with both MPI and Pthread is used to parallelize the short-range molecular dynamics with three parallel algorithms, which are spatial decomposition, atom decomposition and force decomposition, on the common SMP clusters. Thus the computation is parallelized in a two-level manner with Message-Passing among nodes and Shared-Memory within node. The computational results show that the two-level parallel model has different efficiencies with different parallel algorithms compared with only Message-Passing. But in the whole, it can help to improve the computation power by making use of more computing resources.

关 键 词：近程作用 分子动力学模拟 并行计算 MPI 集群结构 

分 类 号：O641.3[理学—物理化学]