Reconstruction of Erasure Correcting Codes for Dependable Distributed Storage System without Spare Disks

Demand is increasing for large-scale dependable storage systems for use in data-intensive servers, such as database and multimedia servers. Conventional dependable disk arrays utilize erasure correcting code to recover lost data stored on failed disks, and require two types of redundant disks: check disks to store check bits, and spare disks to replace failed disks. To reduce the number of redundant disks, this paper proposes a distributed storage system that does not require spare disks. Specifically, the proposed system replaces a failed disk with a check disk rather than with a spare disk. Erasure correction capability of the system gradually degrades with accumulation of failed disks. This paper proposes a reconstruction method for the parity-check matrix of low-density triple-erasure correcting code. Evaluation of the mean time to data loss (MTTDL) shows that, for a storage system with 120 information disks, the proposed storage system with 21 check disks gives an MTTDL of 7.9 x 10⁴ years, while a conventional system with 26 redundant disks gives an MTTDL of 5.8 x 10¹ years, where the mean time to failure of each disk is 0.5 x 10⁶ years and the maintenance interval of each system is 5,376 hours (i.e., 32 weeks).