Research And Implementation Of Reconstruction Optimization For SSD Arrays Based On Popular Data Backup

NAND-flash memory based solid state drive (SSD) has the advantage of high performance, low power consumption, high reliability and excellent shock-resistance. With SSD’s price decreasing and its capacity increasing, it must be more widely used in the future; however, it is easier to trigger a larger number of raw flash memory errors and degrade SSD’s reliability. Besides, the performance of single SSD can not compete for the performance of RAID, as a result, we need to research and design a new SSD array based on RAID technology. When multiple such reliability degraded SSDs organized in a RAID structure SSD failures could occur. In addition, to the best of our knowledge, very little research about RAID structured SSD arrays data reconstruction has been reported in the literature. Thus, a data reconstruction mechanism that can quickly recover the data of a failed SSD onto a replacement SSD is badly needed. The main research work of this thesis is as follows:(1) This thesis describes the research background of data reconstruction strategies for SSD arrays and the development situation of SSD arrays technology, and it also introduces and the development of RAID technology and the current research situation at home and abroad of data reconstruction methods.(2) In this thesis, we propose a new data reconstruction strategy called PDB (Popular Data Backup) for RAID4, RAID5 and SPD-RAID4 (Splitting Parity Disk-RAID4, a variant of RAID4 with multiple parity SSDs) based on SSD. PDB divides each data SSD into a large user zone and a small mirroring zone. While the user zone serves user I/O requests; the mirroring zone of a data SSD backups its immediate neighbor data SSD’s popular read data in real-time and makes data SSDs help each other. Essentially, PDB collaboratively backups popular data among data SSDs to achieve a shorter "window of vulnerability" so as to improve the reliability of storage system.(3) In order to detect the effectiveness of PDB strategy, we transplant the traditional stripe oriented reconstruction (SOR) method, a data reconstruction method which is commonly seen on RAID based on HDD, into SSD arrays based on RAID technology and then make an experiment on SOR based RAID and PDB based RAID. Experimental results demonstrate that compared with the traditional stripe oriented reconstruction (SOR) method, PDB can shorten data reconstruction time up to 31.3%.