作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

›› 2004, Vol. 30 ›› Issue (16): 98-100.

• 论文 • 上一篇    下一篇

MPI容错机制的研究

崔丽青,徐炜民   

  1. 上海大学计算机工程与科学学院,上海大学计算机工程与科学学院 上海 200072
  • 发布日期:2004-08-25

Research of Fault-tolerance Mechanism for MPI

CUI Liqing, XU Weimin   

  • Published:2004-08-25

摘要: MPI是广泛应用于集群系统的并行程序开发环境,MPI的容错是集群系统可靠性的关键问题。该文讨论了MPI标准中的容错,结合协调设置检查点和同步卷回等机制设计了基于检查点的卷回恢复系统MPIChaRR,该系统应用干Linux集群机,MPICH应用程序运行中的节点故障恢复是对用户透明的。

关键词: MPI, 容错, 检查点, 卷回

Abstract: MPI is a parallel programming environment widely used for Cluster. Fault-tolerance of MPI is a key problem to reliability of cluster. This paper discusses the fault-tolerance, MPI standard and presents a checkpoint-based rollback recovery system for Linux cluster, MPIChaRR , with mechanisms of coordinated checkpointing and synchronized rollback. The recovery of node fault during the running of MPICH program is user-transparent.

中图分类号: