基于MPI和OpenMP的三维弹性波方程混合并行有限差分算法

A hybrid parallel finite difference algorithm for 3D elastic wave equation based on MPI and OpenMP

三维弹性波方程有限差分模拟具有大计算量和大内存消耗的特点,在常规计算机上使用传统算法往往无法满足计算要求。该文以高性能计算机集群为平台,基于MPI和OpenMP混合编程技术,构建了一种新型三维弹性波方程并行有限差分算法。该算法基于MPI将总任务分配给多个进程,同时在每个进程中基于OpenMP将子任务分配给多个线程。各个进程具有独立的内存空间,各个线程共享所在进程的内存空间。充液井孔声场的数值模拟结果表明,与基于OpenMP的并行有限差分算法相比,基于MPI和OpenMP的混合并行有限差分算法可以利用计算机集群的多个节点进行并行计算,既极大地提高了计算速度,又有效地降低了单个节点的内存消耗。

The finite difference simulation of 3D elastic wave equation has the characteristics of large computing amount and large memory consumption. It is almost impossible to complete computing tasks by using traditional algorithms on conventional computers. Based on a high performance computer cluster, we propose a new parallel finite difference algorithm for 3D elastic wave equation with the MPI and OpenMP hybrid programming technology. Based on MPI, this algorithm distributes the total task into multiple processes. At the same time, based on OpenMP, this algorithm distributes the subtask of each process into multiple threads.Each process has its independent memory space, and each thread shares the memory space of the process.Numerical simulation results of a fluid-filled borehole model show that compared with the parallel finite difference algorithm based on OpenMP, the hybrid parallel finite difference algorithm based on MPI and OpenMP can use multiple nodes of the computer cluster to complete computing tasks in parallel. Therefore, the hybrid parallel algorithm can greatly improves the computing speed and effectively reduces the memory consumption of a single node.