摘要
针对光滑粒子动力学主要计算量是近邻粒子搜索这一特点,提出了一种基于粒子分解的光滑粒子动力学并行计算方案。利用该方案可以方便地将任意串行光滑粒子动力学代码并行计算,而且每一个时间步内的信息传递量只和粒子总数有关,而和粒子的分布无关,因而特别适合于自由表面流动等大变形问题的并行数值模拟。对一个粒子总数为40万的三维溃坝问题的模拟结果表明:此方案能达到的最大加速比约为16,这一结果可能比空间分解方案(不考虑动态负载均衡)更优。
For the main computation of smoothed particle hydrodynamics was finite nearest particle search, a novel scheme to parallelize the smoothed particle hydrodynamics method based on the concept of particle decomposition was proposed. Any serial smoothed particle hydrodynamics code could be easily parallelized by using the proposed scheme. The amount of information, which was transformed in each time step, depended only on the total number of particles, but not on the spatial distribution of particles. Therefore, the proposed scheme was particularly useful for the parallel simulation of cases involved violent free surface movements. From the simulation results of a 3D dam break case with a total number of 0. 4 million, the proposed scheme can achieve a speedup ratio about 16, which proves that the proposed scheme maybe is better than the domain decomposition scheme (without considering dynamic load balance).
出处
《国防科技大学学报》
EI
CAS
CSCD
北大核心
2015年第4期71-76,共6页
Journal of National University of Defense Technology
基金
国家自然科学基金资助项目(61221491
61303071)
广州科信局基金资助项目(134200026)
关键词
光滑粒子动力学
并行方案
粒子分解
空间分解
smoothed particle hydrodynamics
parallel scheme
particle decomposition
domain decomposition