首页 | 本学科首页   官方微博 | 高级检索  
     

基于相互独立检查点的MPI消息日志系统
引用本文:庞丽萍,陈宝利. 基于相互独立检查点的MPI消息日志系统[J]. 华中科技大学学报(自然科学版), 2004, 32(8): 57-59
作者姓名:庞丽萍  陈宝利
作者单位:华中科技大学,计算机科学与技术学院,湖北,武汉,430074;华中科技大学,计算机科学与技术学院,湖北,武汉,430074
基金项目:国家高技术研究发展计划资助项目 (2 0 0 2AA1Z2 1 0 2 )
摘    要:提出了一种新的MPI消息日志机制及实现原理.它基于发送方的混合日志协议,采用收消息和发消息的全监管机制,使每个进程的消息收、发过程与检查点操作时机相对独立.当一支进程失效时,只回滚该进程本身,减小了因为单支进程失效给整个执行过程带来的进度影响,也使得并行程序具有类似于独立运行程序的自由度.出错过程的相对独立也为同时容多支进程出错提供了,前提.

关 键 词:MPI  并行计算  消息日志  检查点  容错
文章编号:1671-4512(2004)08-0057-03
修稿时间:2003-10-20

The system of MPI message logging based on relatively independent checkpoint
Pang Liping Prof., College of Computer Sci. , Tech.,Huazhong Univ. of Sci. , Tech.,Wuhan ,China. Chen Baoli. The system of MPI message logging based on relatively independent checkpoint[J]. JOURNAL OF HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY.NATURE SCIENCE, 2004, 32(8): 57-59
Authors:Pang Liping Prof.   College of Computer Sci. & Tech.  Huazhong Univ. of Sci. & Tech.  Wuhan   China. Chen Baoli
Affiliation:Pang Liping Prof., College of Computer Sci. & Tech.,Huazhong Univ. of Sci. & Tech.,Wuhan 430074,China. Chen Baoli
Abstract:The infrastructure of MPI message logging was introduced by adopting a logging system relatively independent from each other. The sender based message logging with pessimistic protocol was improved to tolerate more than one failure. The burdensome task of maintaining the consistent system state that each progress should be rollback all independently when a failure occurred was abstained. This can reduce the totally running time of the global task and make the progress have the disengagement approach to independent programs.
Keywords:MPI  parallel computing  message-logging  checkpoint  fault-tolerant
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号