【摘要】 基于单容错编码的数据布局已经不能满足存储系统对可靠性越来越高的要求,而现有的基于多容错编码的数据布局普遍存在编码解码复杂,计算负载大等缺点。文中设计了一种基于双奇偶校验的双容错数据布局DP-RA ID。该布局只需要在磁盘数量为素数的RA ID 5阵列系统中增加1个校验磁盘,通过水平和对角方向的双重奇偶校验实现双容错。该布局的双重校验条纹长度相同,结构简单,易于实现。而且该布局的编码解码算法简单,计算负载小。研究表明该布局与RA ID 5相比,写操作性能会有少许的下降,读操作性能不变,而可靠性大幅度提高。更多还原

【Abstract】 Purpose.Recently there are data placements such as RAID6,DH etc that satisfy the increasing demand for large storage system at the cost of heavy computing load.We propose a DP-RAID(Double Parity-Redundant Arrays of Inexpensive Disk) data placement that is good for large storage system but without increasing computing load.In the full paper,we explain in detail our DP-RAID data placement;here we give just a briefing.DP-RAID data placement,like RAID6 or DH data placement,can tolerate two errors,which are the absolute minimum needed to cope with present-day large storage system which single-error correcting code can no longer cope with.DP-RAID is based on horizontal parity and diagonal parity.This data placement scheme only needs placing a parity disk into a RAID5 disk array system where number of disks is prime number and using double parity to tolerate two simultaneous disk errors.The length of parity stripe of double parity is uniform and easy to implement.In addition,the encoding and decoding algorithm of the data placement scheme is simple and computing load is light.Fig.3 gives the comparison of reliabilities attained respectively by DP-RAID and RAID5(single-error tolerant);Fig.3 shows the variation of reliability with number of disks in the range of 40~320 disks and shows preliminarily that reliability of DP-RAID is much higher than that of RAID5.Table 1 gives four mathematical expressions that show that:(1) DP-RAID is the same in efficiency as RAID5 in reading operations;(2) DP-RAID is slightly inferior to RAID5 in efficiency in writing operations.更多还原