基于OpenMP的湍流场中颗粒碰撞聚合的并行数值模拟

OpenMP-based Parallel Numerical Simulation of Collision and Aggregation Among Particles in Turbulent Field

为了快速求解10000个粒子的湍流碰撞聚合问题,采用OpenMP对Smoluchowski方程的FORTRAN求解程序进行了并行处理.数值结果表明:在不改变串行程序结构的情况下,仅对循环体部分进行并行处理,并行效率可高达80%,且串行程序与并行程序的计算结果完全吻合.对于大计算量循环体的并行计算,采用全部处理器进行并行计算时耗时最小.但是对于小计算量循环体的并行计算,采用全部处理器进行并行计算时耗时不一定最小.

To speed up the simulation of the turbulence-collision-aggregation among 10, 000 particles, OpenMP is introduced to parallelize the FORTRAN source codes for Smoluchowski equation. Numerical results showed that if the serial program＇s structure remains unchanged and only the loops are parallelized, the parallel performance will be up to 80 %, and the computational results of both parallel and serial programs coincide. For the heaw-load computation of loops, the computing time required is minimum if all CPUs are used, but for the light-load one, it is unsure whether the computing time required is minimum even if all CPUs are used.