【作者】 王少华；

【导师】 冯玉才；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 数据库管理系统会因一些不可预知的软硬件故障影响事务的正确执行甚至破坏数据库,降低了系统的可靠性和可用性。数据库系统的备份与恢复部分正是要解决这些问题。本课题以“大型通用数据库管理系统DM V5.6研发及应用”为背景,系统深入地研究了数据库的备份与恢复技术,并改进了达梦备份与恢复子系统,以提高达梦数据库的可靠性和可用性。在现有达梦数据库备份和恢复子系统的基础上设计并实现了备份和日志多文件、异步检查点、基于时间点的恢复和并行恢复。备份多文件是一次备份可根据需要产生多个备份文件,使得备份文件的转移变得容易。日志多文件是指日志系统采用多个日志文件循环使用,避免了系统单个日志文件不断增长的问题。异步检查点采取对脏页上S锁,脏页刷盘完毕立即释放锁的策略,来实现对数据缓冲区的刷盘,使得检查点期间,事务能照常更新。基于时间点恢复采用在系统中添加时间类型日志,并在物理事务结束时,写入时间类型日志来标识物理事务结束时间的方法,只重做指定时间点以内的物理事务,使数据库恢复到时间点的一致性状态。并行恢复采取基于页号取余的日志记录分配算法,一个日志读取线程多个日志重做线程的框架,使得对一个数据页的日志重做由单个线程完成,保证了数据库的一致性。多个日志重做线程并行执行,充分利用了系统的I/O能力,缩短了系统恢复的时间。实验表明异步检查点能提高系统的吞吐量,缩短响应时间。并行恢复能利用系统的I/O并行能力,缩短恢复的时间。更多还原

【Abstract】 The unpredictable software and hardware failures can influence the implementation of transactions or even undermine the database in database management system, and reduce its reliability and availability. Backup and recovery subsystem is to solve these problems. Subject to the“large-scale development and application of generic database management system DM V5.6”as background, we study the database backup and recovery technology and improve backup and recovery subsystem to improve the reliability and availability.Based on the existing DM database backup and recovery subsystem we work on the four main areas, including backup and log multi-file, asynchronous checkpoint, time-based recovery and parallel recovery. Backup multi-file is that a backup operation can produce a number of backup files if need, making the transfer of backup files easier. Log multi-file uses two or more log files circularly to avoid the problem of ever-increasing individual log file.When setting asynchronous checkpoint, we S lock a dirty page, flush the dirty page, then release the S lock after the flushing finished. The transactions can also update pages in the interval of asynchronous checkpoint.Time-based recovery creates a time type log and inserts a time type log in the end of a physical transaction to mark the end time of the physical transaction. When using time-based recovery, we only need to redo the physical database whose time is less than the time point, and back to consistency condition at the time point.Parallel recovery is based on page number mode dispatching algorithm and using the framework of one log-reading thread and multi redo thread, ensuring the log records on one page redoed by a thread, multi redo threads run parallelly. It can make full use of the I/O capability of the system to shorten recovery time.Experiments show asynchronous checkpoint can improve the throughput and shorten response time. Parallel recovery can use the ability of system I/O to shorten recovery time.更多还原