Redundant Disk Arrays: Reliable, Parallel Secondary Storage

摘要: During the past decade, advances in processor and memory technology have given rise to increases computational performance that far outstrip of secondary storage technology. Coupled with emerging small-disk technology, disk arrays provide cost, volume, capacity current subsystems but, by leveraging parallelism, many times their performance. Unfortunately, small disks may much higher failure rates than single large they replace. Redundant Arrays Inexpensive Disks (RAID) use simple redundancy schemes high data reliability. This dissertation investigates encoding, performance, reliability redundant arrays. Organizing into a array is treated as coding problem this dissertation. Among alternatives examined, codes parity are shown effectively correct single, self-identifying failures. The advantages striping across multiple reviewed For transfers parallelism reduces response time. Striping also automatically distributes independent, accesses increase throughput. evaluates lost maintenance data. loss negligible for but can be significant writes because aggregate service time. Because include protect against caused numbers components, it crucial failures characterized. provides evidence lifetimes modeled exponential random variables. Building on an model lifetimes, presents analytic models disk-array lifetime, these event-driven simulation, applies them example array. These incorporate effects independent dependent (shared support hardware) well on-line spare disks. array, show 10% overhead N + 1-parity encoding plus spares 100% conventional mirrored