HV-Recovery:一种闪存数据库的高效恢复方法

HV-Recovery:A High Efficient Recovery Technique for Flash-Based Database

和磁盘相比,闪存作为一种新型的存储设备,具有读写速度快、抗震、省电、体积小等优点.因此,当前的研究普遍认为闪存将取代磁盘成为新一代的数据库二级存储设备.但是,由于闪存具有和磁盘不同的一些固有的读取特性,将当前基于磁盘设计的数据库直接移植到闪存上时,并不能充分发挥闪存设备的优越性.在数据库的恢复过程中,由于闪存的异地更新和重写之前先擦除的特性将带来大量高代价的小的随机写,直接使用传统的恢复方法在闪存数据库中就更难以充分利用闪存的优越性.因此,文中提出了一种对闪存中天然存在的数据的历史版本来进行管理和利用的恢复方法HV-recovery,来改进undo恢复的性能.通过和开源数据库Oracle Berkeley DB的比较,实验结果表明HV-recovery是原有的恢复算法性能的2~8倍,充分说明了其优越性.

Flash memory,as a new kind of data storage media,has a lot of attractive characteristics when compared with Hard Drive Disk（HDD）such as fast access speed,shock resistance,power saving,lighter form and low noise.Therefore flash memory is considered as the main storage device instead of disk in the next generation.However,traditional disk-based database can＇t take full advantage of high I/O performance of flash memory if we transfer it to flash memory without modification.The main reason is flash memory embraces different access characteristics with HDD.As for recovery,the situation becomes more serious because the out-of-place update model and erase-before-rewrite of flash memory lead to high cost of large quantity of minor random writes during the course of recovery.This paper proposes a recovery method,HV-recovery,to improve the performance of undo.HV-recovery makes use of the history versions of data which is naturally emerged in flash memory due to the out-of-place update.Experimental results on Oracle Berkeley DB show that the HV-recovery outperforms traditional recovery in 2×~8×.The results demonstrate the high efficiency of the method.