超步透导的回卷恢复  

Superstep-Induced Rollback Recovery

作　　者：丁俊[1] 童维勤[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200072

出　　处：《小型微型计算机系统》2002年第6期731-735,共5页Journal of Chinese Computer Systems

基　　金：上海市教育委员会曙光计划资助项目

摘　　要：工作站机群系统已成为分布式并行处理发展的主流方向之一 .随着机群系统应用领域的逐渐拓展和规模的不断扩大 ,人们对其可靠性的要求日益提高 .设计高可靠的群机系统 ,需要着重研究其系统容错技术 .本文叙述了并行异构环境回卷恢复和检查点派生 .实现透明的可移植容错和负载均衡能力 .避免调整检查点就可以构成全局一致性状态 .不仅使 BSP应用程序自治容错能力 ,而且能够在机群 (Clusters)间迁移 ,保持系统负载均衡 .重点介绍检查点设置、检查点派生、卷回。Cluster of Workstations(COW) now becomes one of the leading teachnologies in the field of distributed parallel processing.To implement the COW with high reliability,it is necessary research its system reliability technique.This paper discusses techniques of checkpointing ,checkpoint propagation and rollback recovery which enable parallel programs written using the BSPlib communication library to tolerate fault and to be ported. The techniques can be used in a mode which is almost totally transparent to the programmer. The critical techniques of checkpointing?checkpoint propagation ?rollback recovery and process migration are focused on in user level.

关 键 词：超步诱导 回卷恢复 容错 故障恢复 进程迁移 进程状态 工作站机群系统 

分 类 号：TP368.5[自动化与计算机技术—计算机系统结构]