基于MPI的三维枝晶生长相场法的并行计算

Parallel computation of three-dimensional dendritic growth with MPI-based phase-field method

充分利用MPI(message passing interface)在并行环境下远高于单CPU的强大计算能力,探索基于MPI的并行系统结构,求解三维枝晶生长的高性能计算方法.通过多进程的并发执行,实现三维相场方程求解的并行计算,探讨MPI中点对点通信与集合通信在并行计算时数据传输的效率,讨论热噪声幅值Fu=0与Fu=10-3时三维枝晶生长过程.计算结果表明:基于MPI的并行算法可使模拟尺度达到1 000×1 000×1 000网格,大大提高可模拟尺度;采用集合通信模式比点对点通信模式具有更高的并行效率,更加适合大规模并行计算环境.

The powerful MPI（message passing interface）-based computation capacity in parallel environment,which is much higher than the single CPU,is employed sufficiently to explore a high-performance computation method based on MPI parallel system structure for finding the solution of three-dimensional dendrite growth.The parallel computation for finding the solution of three-dimensional phase-field equation is implemented with concurrent execution of multiple processes,the efficiency problem of data transmission in parallel computation of point-to-point communication and collective communication with MPI is explored,and the three-dimensional dendritic growth process is also discussed when thermal noise amplitude FU=0and FU=10-3.The calculation results show that the MPI-based parallel algorithm can make the simulation scale to amount to 1 000×1 000×1 000,being greatly increased.The collective communication mode will have higher parallel efficiency compared to the point-to-point communication,and higher suitability for large-scale parallel computing task environment.