节点文献
基于ISCSI连续数据保护系统的研究和实现
Research and Implementation of Continuous Data Protection System Based on ISCSI
【作者】 喻强；
【导师】 鞠大鹏；
【作者基本信息】 清华大学 ， 计算机科学与技术， 2008， 硕士
【摘要】 随着计算机在社会生活中的普及,存储的数据越来越成为人们的宝贵财富。对数据安全的威胁主要有硬件的错误和人为因素导致的软错误。RAID技术采用硬件控制器对磁盘数据自动做校验值,当某个磁盘损坏时可以利用这些校验值把损坏的数据恢复出来,它只能处理硬件的错误。备份技术可以处理软错误,但是由于备份是周期性的,无法做到完全的数据无损失。CDP(Continuous DataProtection)是一种新的数据保护技术,它主要是用来处理软错误的,并且可以做到完全无数据损失。CDP既可以用硬件实现也可以用软件实现,本文就是从软件实现的角度对CDP做了研究,指出其中的关键技术,并且对性能做了详尽的分析。本文主要的研究工作和贡献有:1.明确CDP的应用场景和功能界限。CDP是对传统的备份技术的补充,它弥补了用备份来恢复数据带来的时间长、数据有损失等缺点。2.设计并实现了Log-Struct结构的磁盘存储方式,并且对这种存储方式的性能进行了详尽的分析。3.分级数据读取结构的设计。把数据的读取分为好几个层次,不同的层次性能相差很大,通过实现高的命中率来提高系统的总体读取性能。4.对CDP系统的空间占用情况、读取性能、写入性能进行了详尽的分析,使得CDP的性能有了理论上的支持。
【Abstract】 With the widespread use of computer in social life, the data become a great treasure for people. Hardware failure and soft errors which were caused by human are two major threats to data security. RAID use a controller to calculate disk data parities automatically, if one disk fails, RAID can use these parities to recover data. It can only cover the hardware failure threat. Although backup can deal with soft errors, we can not avoid data loss since backup always be done in period. CDP is a new technique for data protection. It is mainly used to cover the soft errors, and it can make sure there is no data loss. CDP can be implemented either by hardware or software. In this paper, we do some research on software implementation for CDP. We point out the key techniques during the implementation and at the end analyze the performance in details.The research work and contribution in this paper are:Make the applications and functions of CDP clear. CDP is a supplement to the traditional backup technique. It solves problems in backup technique that it would take a long time to restore the data and there still have some data loss.Design and implement the disk storage based on Log-Struct infrastructure, and analyze the performance in details.The design of data read hierarchy. We divide the read process into several levels. Different levels differ from their read performance. We achieve a high performance for the whole system through a high hit rate in the upper levels of read.Analyze the space occupation, read and write performance of CDP, the result gives a support to performance analysis in theory.