Candidate MDS array codes for tolerating three disk failures in RAID-7 architectures

Abstract

Current storage systems use RAID-5 and RAID-6 architectures to provide protection against one and two disk failures, respectively. However, as the size of storage system grows rapidly three concurrent disk failures are becoming more frequent. To cope up with three disk failure, we propose a new RAID level, i.e., RAID-7, for which three-column-erasure tolerating MDS array codes are needed. However, it is an open question as to which MDS array codes should be used for RAID-7. In this paper, we compare different array codes, which can be used in RAID-7 systems that require storage efficiency (the ratio of number of information symbols to encoding (or codeword) symbols) <= 0.5. The paper discusses three-column-erasure tolerating MDS array codes proposed in the literature namely, [5,2] 2 x 5 BP-XOR code, [6,3] 4 x 6 lowest-density array code, [6,3] 2 x 6 STAR code, [6,3] 4 x 6 generalized RDP code. The paper introduces a new three-column-erasure tolerating [6,3] 2 x 6 almost BP-XOR codes. We analyze annual failure rate, storage efficiency, worst case normalized encoding/update/repairing/read complexity, repair bandwidth, and number of buffers required for these codes. We also provide experimental results to understand the average case encoding and repairing complexity of BP-XOR, STAR, GRDP, and almost BP-XOR codes by implementing them in software. From our analysis and experimental results, we conclude that [6,3] 2 x 6 almost BP-XOR are best suited for RAID-7 systems with storage efficiency <= 0.5.