Evaluation of the impacts of data hot-spots in disk arrays on performance and availability

This paper investigates the impacts of data hot spots on performance and system availability of disk arrays. We perform experiments on RAID 5 to show quantitatively how hot-spots affect the performance in terms of I/O throughput and availability in terms of mean time to repair, MTTR. We find that for small access workloads, the performance, and system availability, is dominated by the access pattern (read/write), with negligible change due to increases in hot-spot activity. Workloads with larger accesses see a much larger performance and availability impact with increased hot-spot activity.