MPI + OpenMPI混合编程的实现与性能分析

Implementation and Performance Analysis of MPI + Open MPI Hybrid Programming

摘要: 为充分发挥SMP集群的计算能力，结合SMP集群并行系统体系架构的特点，采用MPI + OpenMP混合并行编程模式，研究了基于SMP集群的多级混合并行程序的设计和实现方法。通过节点间使用MPI消息传递接口实现任务的粗粒度划分，节点内使用OpenMP实现细粒度任务并行划分的方法，开发了混合并行编程性能测试用例，并对其进行了测试和分析。测试结果表明，MPI + OpenMP混合编程模式，能够有效利用SMP集群节点间和节点内多级并行机制，充分发挥消息传递编程模型和共享内存编程模型各自的优势，可有效提升SMP集群的计算效率。

Abstract: In order to give full play to the computing power of SMP cluster, combined with the characteristics of SMP cluster parallel system architecture, the design and implementation of multilevel hybrid parallel program based on SMP cluster are studied by using MPI + OpenMP hybrid parallel programming mode. The performance test case of hybrid parallel programming is developed by using MPI to realize the rough degree partition of tasks between nodes and OpenMP to realize the fine degree partition of tasks in nodes. The test result shows that MPI + OpenMP hybrid programming mode can effectively make use of the multi-level parallel mechanism of SMP cluster, and give full play to the advantages of the two programming models and effectively improve the computing efficiency of SMP cluster.