基于数据块的数据保护技术研究

发布时间：2018-01-02 11:40

本文关键词：基于数据块的数据保护技术研究　出处：《华中科技大学》2013年博士论文　论文类型：学位论文

【摘要】：随着社会对信息化依赖程度越来越高,数据丢失往往会造成巨大的损失。数据丢失的原因主要可以分为两类,一类是系统的软硬件故障,另一类则是人为误删除或非法篡改。系统的意外停机会直接导致系统缓存中数据丢失。即便部分缓存采用的是非易失性存储器,也可能因其缓存映射表和状态信息异常失效而依然导致整个缓存不可使用。为了抵抗这种风险需要缓存有可持续性。人为操作失误或篡改则会直接改变当前数据存储状态,为了抵抗这一风险也需要存储系统有在时间轴上向前回滚的能力。首先为解决缓存系统发生意外时数据丢失的问题,设计了一个基于固态盘(SSD)的缓存系统PFC (Persistent Flash Cache)。PFC采用改进的组相联映射地址映射方式,在节省元数据开销的同时,利用局部性原理加快了数据写回硬盘的速度。同时PFC通过将部分存在易失性主存(RAM)中的缓存系统元数据映射到非易失性的固态盘中并实时更新的方式,保证缓存中的数据在经历计划停机或意外停机时都不丢失。其次提出了低开销高可用元数据更新方法LUFUW(Lazy-Update Following an Update-Write)。LUFUW通过借鉴计算机网络技术中TCP协议确认帧是通过数据帧捎回这一策略,使某个元数据条目的更新可以通过同一个元数据块中的其它元数据条目的更新捎带回固态盘缓存的元数据区。从而在只增加少量元数据更新开销的情况下,保证系统重启后,固态盘中绝大部分缓存数据可用。实验结果表明在非正常关机启动时,相对于传统的只有14%缓存数据可用元数据更新方式,新的元数据更新策略使99%以上缓存数据可用。从而使系统重启后缓存的性能是原始Flashcache的两倍。最后设计了一个在时间轴上有向前回滚能力的高效连续数据保护系统ST-CDP (Snapshot in TRAP Continuous Data Protection)。传统的连续数据保护系统为了保存系统中每一次修改,会产生大量的记录数据。ST-CDP通过引入异或运算提高了数据的压缩比,有效的节省存储空间。通过结合传统的快照技术和现有的连续数据保护技术,在记录数据变化的校验链中周期性的插入快照,使系统能确保在发生故障时进行快速、可靠的数据恢复。通过建立如何插入、何时插入快照的数学模型,优化存储空间和恢复时间。基于此数学模型,在Linux内核上设计并实现了此连续数据保护系统。实现表明,ST-CDP和传统的连续数据保护技术在运行时开销相似,但其连续数据保护日志空间非常开销小,接近于即时任意点数据恢复(TRAP)。
[Abstract]:With the increasing dependence of society on information technology, data loss will often cause huge losses. The reasons for data loss can be divided into two categories, one is the system software and hardware failures. The other is human error deletion or illegal tampering. The unexpected shutdown of the system will directly lead to the loss of data in the system cache, even if part of the cache uses non-volatile memory. It is also possible that the entire cache is not usable because its cache mapping table and state information are not working. To resist this risk, cache sustainability is required. Human error or tampering directly changes the current data. Storage status. To resist this risk, storage systems are also required to roll back on the timeline. First of all, to solve the cache system unexpected data loss problem. A cache system, PFC persistent Flash Cache).PFC, based on SSD, is designed in this paper. It uses improved group associative mapping address mapping method. While saving metadata overhead. The local principle is used to speed up the speed of data being written back to the hard disk. At the same time, PFC uses part of the volatile main memory (RAM) to write the data back to the hard disk. The way in which cache system metadata is mapped to a non-volatile solid-state disk and updated in real time. Ensure that data in the cache is not lost during scheduled or unexpected downtime. Secondly, LUFUW(Lazy-Update Following an Update-Write-based method with low overhead and high availability is proposed. LUFUW uses the TCP protocol in computer network technology to confirm that frames are sent back by data frames. An update of a metadata entry can be piggyback to a cached metadata area of a solid-state disk through an update of other metadata entries in the same metadata block, thereby adding only a small amount of metadata update overhead. After the system is restarted, most of the cache data are available in the solid state disk. The experimental results show that only 14% cache data can be updated by metadata in the abnormal shutdown startup. The new metadata update strategy makes more than 99% cache data available, making the cache performance twice as good as the original Flashcache after the system restarts. Finally, an efficient continuous data protection system, ST-CDP, which has the ability to roll back forward on the time axis, is designed. Snapshot in TRAP Continuous Data protection. Traditional continuous data protection system to save every change in the system. ST-CDP will produce a large amount of recorded data. ST-CDP can improve the compression ratio of data by introducing XOR operation and save storage space effectively by combining the traditional snapshot technology and the existing continuous data protection technology. The periodic insertion of snapshots in the check chain that records changes of data enables the system to ensure rapid and reliable data recovery in the event of failure. A mathematical model of how to insert and when to insert a snapshot is established. Based on this mathematical model, the continuous data protection system is designed and implemented on the Linux kernel. The overhead of ST-CDP is similar to that of the traditional continuous data protection technology, but its continuous data protection log space is very small, which is close to the real-time data recovery at any point.
【学位授予单位】：华中科技大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP333

【共引文献】