广域网环境下多主机文件同步技术研究  

Snooping-Balanced Cache Coherence Protocol for Chip Multiprocessors

作　　者：李强[1] 朱立谷[1] 曾赛峰[1] 

机构地区：[1]中国传媒大学计算机学院,北京100024

出　　处：《高性能计算技术》2012年第5期36-40,共5页

基　　金：本文获国家科技支撑计划（2011BAH04B05）的资助.

摘　　要：对广域网上多个节点进行文件同步时，因节点频繁地加入与离开使得直接的点对点同步不可行。对文件的更新发现，单一地采用状态比对和追踪文件变化过程都不可取，前者因频繁计算副本状态资源消耗过大，后者需要保存的信息量会无限增长。在应用其它节点更新时，为获得更高的可用性，不能阻断用户对被同步文件的访问，从而引入了一种新的冲突类型。本文遵循乐观复制算法，采用星型结构连接副本节点，融合了状态和操作传输，实现了一个节点可灵活扩展的文件同步系统，改进了基于状态比对的更新发现，并结合文件变化日志共同发掘文件更新。最后，通过语义层的操作转化解决了同步文件被用户占用的访问冲突。Directly point-to-point file synchronization for the nodes distributed in wan is not feasible because the nodes continually join and leave the network. For the exploring of file update, separately using state-based file comparison or tracking file changes is not desirable. For the former, the continual calculation of the replication state consumes too much resource. For the latter, the amount of information needed to save will grow indefinitely. In order to obtain higher availability, when the other nodes updates, we can not blocking the access of users to the synchronized documents, but then it appears a new type of conflict. This paper follows optimistic replication algorithm, connects replica nodes in star topology, and integrates the state and operation transfer to implement a node-scalable file synchronization system. It also improves the state-based update found and explores file update in conjunction with file change log. At last, we solve the access conflicts on the synchronized files through operation conversion of the semantic layer.

关 键 词：乐观复制算法 更新发现 访问冲突 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]