MPI_ALLGATHER实现算法的改进与性能分析  

Improvement and performance analysis of MPI_ALLGATHER algorithm

作　　者：李占胜[1] 毕会娟[1] 都志辉[2] 焦青[3] 

机构地区：[1]华北计算技术研究所软件平台研究室,北京100083 [2]清华大学计算机科学与技术系,北京100084 [3]鲁东大学图书馆,山东烟台264025

出　　处：《计算机工程与应用》2007年第27期66-71,共6页Computer Engineering and Applications

基　　金：中国科学院计算机网络信息中心超级计算中心支持(Supported by Supercomputing Center;CNIC;CAS)

摘　　要：首先分析了影响MPI组通信性能的各方面因素,提出了一种衡量算法性能的模型。基于这种分析及模型,提出了一种将邻居交换和递归倍增两种算法结合的新的MPI_ALLGATHER实现算法。新的算法比邻居交换算法通信次数少,比递归倍增算法具有较好的通信局部性。通过在高性能机群系统中的测试,发现新算法在多种情况下比邻居交换算法具有更优的性能,在中等长度消息通信时具有最优的性能,在长消息通信时性能比递归倍增算法和Bruck算法的性能更优,且在长消息通信时多数情况下性能最优。We discuss issues related to the high performance implementation of collective communication operations in MPICH for clusters in this paper.The Ring and Neighbor Exchange algorithms have the best communication locality property,and they have more communication steps than others.The Recursive Doubling and Bruck algorithms have less communication steps,and they have worse communication locality property than others.Based on the discussion,we present a new MPI_ALLGATHER algo- rithm that combines the Neighbor Exchange and Recm^ive Doubling algorithms.The new algorithm has less data transfer times than the Neighbor Exchange algorithm and better communication locality property than the Recursive Doubling algorithm.We eval- uate the new algorithm in Myrinet cluster and find that the new algorithm performs better than the Neighbor Exchange algorithm and sometimes performs the best for medium-size messages among all the algorithms.The new algorithm also performs better for long-size messages than the Recursive Doubling and Bruck algorithms.

关 键 词：并行编程 MPI 组通信 MPI_ALLGATHER算法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]