Reliability and performance of disk arrays

Introduction of redundancy into secondary storage systems in the form of Redundant Arrays of Independent Disks (RAIDs) has generated both research activity and commercial applications. The thesis investigates the performance and reliability of RAIDs analytically. We give exact Markov models to derive reliability results. We use queuing network methods to calculate RAID performance in the failure free and post-failure states. We prove the necessity of non-volatile storage through the evaluation of synchronized write protocols. In addition to Level 5 RAIDs with and without spares, we treat higher redundancy extensions of the Level 5 RAID: two-dimensional RAIDs and disk arrays based on Maximum Distance Separable (MDS) codes. We define and investigate a string sensitive declustering scheme, the Almost Complete Address Translation Scheme (ACATS). Finally, we extend the use of MDS codes to implement a storage space saving distributed database.