一种基于动态区域分解的SPH并行算法研究及应用

Domain Decomposition Based SPH Parallel Computing Method Study and Its Application

为提高SPH(光滑粒子流体动力学)方法的计算速度和大规模仿真能力,发展了一种基于动态区域分解的SPH并行算法。采用基于维度的因数分解法实现任意数目的并行计算区域分解。采用二分法实现快速并行切割面搜索。通过改进子区域分配方法,有效减少计算节点间通信数据量。基于该算法开发了并行SPH计算软件PTS,并对弹丸超高速斜撞击平板形成碎片云过程进行了并行SPH仿真。仿真结果表明:该算法适应大规模冲击动力学仿真,可有效保证仿真过程中的负载平衡,并具有较好的线性加速特性。

To improve the speed and the capability of large scale simulation of SPH （Smoothed particle hydrodynamics）, a parallel SPH method based on dynamic domain decomposition is developed. The decomposition of arbitrary numbers of subdomains is achieved by adopting dimension based factorable method. Rapid parallel cut plane search is achieved by adopting bisection method. The data communication between compute nodes is reduced by improving the subdomain assignment. A parallel SPH program was developed based on this method, with which the process of projectile hypervelocity impacting thin plate producing debris cloud was simulated. The result shows that： the method is suitable for the simulation of large scale shock dynamic problem, the load balance is well kept during the simulation, and the method shows a good linear acceleration feature.