| 注册
首页|期刊导航|计算机工程与科学|基于"天河二号"聚合通信卸载特性的MPI_Barrier优化

基于"天河二号"聚合通信卸载特性的MPI_Barrier优化

朱琦 戴艺 彭晋韬 谢旻 梁崇山 刘鹏 杨博 刘杰

计算机工程与科学2025,Vol.47Issue(3):400-411,12.
计算机工程与科学2025,Vol.47Issue(3):400-411,12.DOI:10.3969/j.issn.1007-130X.2025.03.003

基于"天河二号"聚合通信卸载特性的MPI_Barrier优化

Optimization of MPI_Barrier based on the offloading characteristics of Tianhe-2

朱琦 1戴艺 2彭晋韬 1谢旻 2梁崇山 2刘鹏 1杨博 2刘杰1

作者信息

  • 1. 国防科技大学计算机学院,湖南长沙 410073||国防科技大学高端装备数字化软件湖南省重点实验室,湖南长沙 410073||国防科技大学并行与分布计算全国重点实验室,湖南长沙 410073
  • 2. 国防科技大学计算机学院,湖南长沙 410073
  • 折叠

摘要

Abstract

Barrier,as a fundamental operation in message passing interface(MPI)programs,is one of the critical mechanisms ensuring the correct execution of programs.Existing Barrier implementation schemes primarily suffer from two defects:firstly,there is significant redundant data path transmission overhead during inter-node synchronization;secondly,there are numerous cache misses during intra-node synchronization.To address these performance limitations,this paper proposes two optimization techniques tailored for the aggregate communication offload features of the Tianhe-2 customized net-work,TH-Express:Barrier acceleration based on GLEX NIC and shared memory flag bits rearrange-ment.These techniques effectively reduce the synchronization overhead between nodes and improve the synchronization efficiency within nodes based on shared memory.Based on the aforementioned optimiza-tion methods,this paper redesigns the MPIBarrier algorithm and integrates it into the MPI communica-tion library.Performance tests of the proposed scheme are conducted on micro-benchmark programs and real applications running on the National Supercomputing Center in Changsha,with a scale of up to 7168 nodes.Experimental results show that the optimized MPI_Barrier collective operation achieves a speed-up ranging from 1.3 to 14.5 times,and in application-level real-load evaluations,the performance im-provement reaches up to 54%.

关键词

MPI/Barrier/大规模并行应用/NIC聚合通信卸载

Key words

massage passing interface(MPI)/Barrier/massively parallel applications/NIC collective communication offloading

分类

计算机与自动化

引用本文复制引用

朱琦,戴艺,彭晋韬,谢旻,梁崇山,刘鹏,杨博,刘杰..基于"天河二号"聚合通信卸载特性的MPI_Barrier优化[J].计算机工程与科学,2025,47(3):400-411,12.

基金项目

国家自然科学基金(62272476) (62272476)

国家重点研发计划(2021YFBO300101) (2021YFBO300101)

国家自然科学基金重点项目(U22B2005) (U22B2005)

并行与分布处理国家重点实验室基金(2021-KJWPDL-08) (2021-KJWPDL-08)

计算机工程与科学

OA北大核心

1007-130X

访问量1
|
下载量0
段落导航相关论文