华中科技大学学报（自然科学版）Issue(12)：75-80,6.DOI:10.13245/j.hust.141215
基于进程到达模式感知的 MPI 广播性能优化
Performance optimization of MPI broadcast based on process arrival pattern awareness
摘要
Abstract
The unbalanced process arrival pattern (PAP) has a profound impact on the performance of message passing interface (MPI) collective communication ,to address this problem ,an optimized broadcast method was proposed based on PAP awareness .By considering the architecture of multi‐core clusters ,the optimized algorithm was structured on the basis of pipelined broadcast algorithm . There were two PAP awareness mechanisms:intra‐node and inter‐node , thus the earlier arrived processes could participate in the operation as soon as possible .In consequence the latency time was cut down and the communication performance was improved .According to the simulation of theory analysis ,the unbalance PAP has little impact on the optimized algorithm .Then ,the performance of the proposed algorithm was evaluated with different message sizes ,PAPs and process numbers .The results show that the proposed algorithm improves the performance of M PI broadcast under the unbal‐anced PAP effectively ,with speeding up from 129% to 174% .关键词
消息传递接口/进程到达模式/多核集群/集合通信/流水线广播Key words
message passing interface (MPI)/process arrival pattern/multi-core cluster/collective communication/pipelined broadcast分类
信息技术与安全科学引用本文复制引用
吴文波,李林,姚新宇,刘丽丽..基于进程到达模式感知的 MPI 广播性能优化[J].华中科技大学学报（自然科学版）,2014,(12):75-80,6.基金项目
国家自然科学基金资助项目（61374185）；中国洛阳电子装备试验中心电子信息系统复杂电磁环境效应国家重点实验室开放基金资助项目（CEMEE2014K0303B）． （）
