一种基于存储虚拟化的异步远程镜像系统

An Asynchronous Remote Mirror System Based on Storage Virtualization

信息数据在当今社会中的重要性日益提高.远程镜像系统通过配置冗余的硬件和相应的高可靠软件,能够保证信息系统的数据安全.但是,已有的远程镜像系统都部分存在着依赖于专用设备或底层驱动、性能差等不足.该文提出一种应用于存储区域网络环境的、基于存储虚拟化的异步远程镜像系统.首先,设计了异步镜像逻辑卷,作为虚拟化的数据容器,该卷能够自动追踪并异步传输数据更新,节省网络带宽,同时不依赖于任何的底层设备和驱动.其次,提出了一种不间断服务的异步镜像协议,支持服务在镜像主节点和从节点间的无缝迁移,支持灾难发生和灾后恢复时主从节点的自动切换和不间断服务,同时,协议完全在镜像节点内执行,不影响客户主机的性能.最后,针对镜像卷的读写特点,设计了基于锁的镜像卷访问控制机制,既保证数据一致性,又能在此前提下尽可能提高读写并发性.FTPtrace(文件传输服务跟踪记录)的重放实验证明,在人为引入主节点失效的情况下,服务能够无缝迁移;在主节点灾后数据恢复进行的同时,能够保证服务不间断;最后,FTPTrace的重放流程在"主节点失效-服务迁移-灾后恢复"场景下仅比无故障时多花费14%的时间.

Computerized data becomes more and more critical in today＇s digital life. Remote mirror can keep information systems alive with redundant hardware and software, even encountering site corruptions. Unfortunately, existing mirror systems have some limitations such as relying on specific storage devices, poor performance, and so on. This paper presents a storage virtualiza- tion based asynchronous remote mirror system. First, in this system, all data are protected by a logical container, that is, an asynchronous mirror logical volume, which can trace and transfer data updating automatically. This way, there is no device and driver dependency at all. Secondly, this paper presents an advanced non-stop mirror protocol to support seamless service switching when site corruption happens and automatic data recovery at rebuilding process after disaster. In addition, all the protocol can be executed on mirror IO nodes in SAN, introducing little overhead to the protected servers. Finally, it gives design on lock based concurrent access control meeha-. nism for the asynchronous mirror logical volume. With the help of the lock, not only data consis- tency can be maintained, but also good concurrent read/write performance can be achieved. Ex- periments and evaluation results demonstrate that the prototype system has the ability to provide high reliability while only introduce little overhead： tolerated a site corruption while kept provi- ding service, costing only 14% more time.