摘要
扩散系数计算量非常大,为提高计算效率研究采用并行计算替代单机计算:用非阻塞通信替代阻塞通信;利用元胞分解的成果创建要拷贝的外层粒子列表;用拷贝粒子和移动粒子的重叠部分来减少通信量,在空间分解(SD)算法的基础上对MD模拟的并行算法进行了改进,使并行计算的效率提高了28.64%。用改进的SD并行计算方法,以RedHat Linux 8.0位操作系统和基于Message Passing Interface(MPI)消息传递界面的四节点计算机集群系统(PC Cluster)。对水分子在宏量条件下以及受限于硅酸岩狭缝中的速度自相关函数进行了MD模拟计算。结果表明,并行算法可以使速度自相关函数及相应的自扩散系数的MD模拟时间大为缩短。
Spatial decomposition (SD) has been improved from three facets: Blocked communication replaces non-blocked communication, the result of cell decomposition is used, and overlap of copying and removing particle is used in order to reduce traffic. After modification, the program is more 28.64 percent effective. A PC Cluster, which contains four computing nodes, based on MPI ( Message Passing Interface) for molecular simulation has been established with RedHat Linux 8.0 operating system. Velocity auto-correlative function (VACF) of water in macrovolumn and confined in silicate is computed. The result shows that parallel MD simulation can reduce the time consumed in calculating VACF and diffusivity.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2005年第2期85-90,共6页
Computers and Applied Chemistry
基金
国家重大基础研究发展规划(973)项目(G1999043309)
国家自然科学基金项目(20277034)