关键词: 云计算, 磁盘错误, 潜在扇区错误, 纠删码, 磁盘内冗余

Abstract: The development and application of cloud computing set higher requirement for the fault-tolerant capability of the storage systems. Erasure code has been widely used to generate device-level redundancy to protect against device failures, while has less space efficiency when resisting the sector failures. Current optimization schemes for the sector failures only resist the failures of small amounts of the sectors or specific sectors. In this paper, we propose a hybrid coding scheme (intra- and inter-device redundancy, IIDR) combining inter-device redundancy with intra-device redundancy based on the homomorphism property of MDS (maximum distance separable) codes, which employs global parity sector against sector failures in the data disks when adding parity device against device failures, and optimizes the ability to process single-sector errors taking advantage of intra-device coding to generate local parity sectors. In the end, the correctness proof and performance analysis are shown in this paper, and the results indicate that our scheme can protect against device failures and sector failures of any distribution, and the computing cost of recovering single-sector errors is much lower, and the update performance is better. Compared with traditional intra-device coding schemes, our scheme comes with less space usage.

Key words: cloud computing, device failures, latent sector errors (LSEs), erasure code, intra-disk redundancy