一种可扩展分布式RAID存储集群系统

An Expandable Distributed RAID Storage Cluster System

一种可扩展、单位存储价格低廉的分布式RAID存储集群系统被提出,用以取代硬件RAID磁盘阵列.该系统具有3个特性:1)通过网络共享存储模型共享存储资源,每个数据节点处于完全对等的地位,是一种P2P架构的实现,易于扩展;2)在用户空间对节点间的RAID元数据进行同步,具有SingleI/OSpace(SIOS)特性,并能支持各种RAID技术;3)提供内核空间虚拟块设备的访问接口,对文件系统透明.测试结果表明该系统最高连续读带宽达到190MBps;在RAID6冗余配置下,1～2个节点失效不会造成数据访问服务的中断,但1个节点失效会造成连续读带宽下降15%,2个节点失效造成连续读带宽下降18%.

An expandable, low price per storage unit, distributed RAID storage cluster system is presented to substitute the traditional hardware RAID array. It has three important features. First, it＇s a P2P infrastructure via network share storage model, with each data node in a completely equal status, and provides a great convenience in the expandability of distributed RAID system. Secondly, it sustains single I/O space（SIOS）feature by synchronizing RAID operations in user space, which is independent of the kernel-level RAID implementation, thus can support various advanced RAID technologies. Thirdly, it provides a virtual block device access interface, transparent to upper level file system and database system, thus can be directly used by application without any changes. Also, the system provides a node failure detection and recovery mechanism, which can effectively detect and replace failure nodes to guarantee data integrity and maintain system reliability. Experiments show that network shared storage model brings little impact on storage access bandwidth and the maximal sequential read bandwidth can reach 190 MBps, which is limited by the bandwidth of gigabit Ethernet. Also, experiments validate the reliability of the distributed RAID system. Under RAID6 configuration, one or two nodes failure will not interrupt data service, but one node failure leads to 15 % bandwidth decline and two nodes failure leads to 18 % bandwidth decline.