集群存储系统数据容错研究

Research on Fault Tolerance of Cluster-based Storage System

针对当前网络存储技术存在的存储服务瓶颈问题,介绍了一个高可靠可扩展PC集群存储系统的容错设计方案.利用广泛用于加密解密技术的SHA-1算法来产生数据对象的副本ID,然后根据各个存储结点的权值信息,利用散列函数将数据对象复制到各个存储结点,使系统具有很好的容错性.另外描述了系统中对数据对象存储、读取、删除的基本方法,并介绍一种类似基于时戳的法定数选举方法来维护系统中数据对象副本的一致性.最后对系统的可靠性以及系统所支持的最大用户量进行了分析.

Aiming at the bottle problem of storage service in current network storage technology, a fault tolerance scheme of high reliability scalable cluster-based storage system is proposed. The identification of a data object＇s replication is generated by the SHA-1 algorithm which widely used in encryption and decryption techniques. According to the weight information of each storage node, the data object＇s replication is hashed to different storage node, in consequence, the better fault tolerance of the system is achieved. The basic approach to store, retrieve and delete data object is described, as well as a similar method to timestamp-based majority voting method to maintain the consistency of the system＇s data object. Finally, the system reliability and the max users supported by the system are analyzed.