Research On Coding And Fault Tolerance Technology In Disk Array

With the development of massive storage system and its applications in complexenvironments, disk failure already becomes a norm. Thus the research of reliabilityproblem becomes very popular in both academic and industry. To resolve this problem,disks are always organized as disk arrays (RAIDs), using coding and fault tolerancetechnology to provide data reliability. This dissertation focus on the coding and faulttolerance technology, the main contributions include:1. Shifted Element Arrangement in Mirror Disk Arrays: This dissertation pro-vides a shifted element arrangement for mirror method. This method adjust theelement arrangement in mirror disk array, thus minimizing the number of readaccesses in recovery process. When disk numbers equals to3to7, the experi-mental results show that the shifted element arrangement improves the recoveryefciency by a factor of1.54to4.55.2. Efcient RAID-6Code for Arbitrary Size of Disk Array: This dissertationproposes a novel RAID-6code: generalized X-code. The code inherits the opti-mal encoding and updating efciency of RAID-6vertical cod. Then by adjustingvertical code into horizontal form, the code could adapt to arbitrary size of diskarray.3. Erasure Code Architecture for High Fault Tolerance: This dissertation pro-vides a novel erasure code architecture with high fault tolerance: nested code.This architecture considers erasure codes on two diferent dimension, nestingone code on another code. Thanks to the orthogonal protection, the new codearchitecture provides high fault tolerance.4. Load-Balance Recovery Schemes for Single-disk Failure Problem of AnyErasure Code: This dissertation discusses two recovery schemes for single-diskfailure problem of any erasure code. The first scheme provides load-balanceunder the condition of minimizing total read amount, while the second one min- imizes the maximum read amount from a single disk, thus providing more load-balance. The goal of both two schemes are load-balance by minimizing the num-ber of read accesses, thus improving the recovery efciency. In an array with16disks, the experimental results show that these two schemes improve the recov-ery speed by as high as15.5%and19.9%, comparing with the state-of-the-artscheme.