作者: Gregory R. Ganger , Matthew Wachs , Michael Abd-El-Malek , Eno Thereska
DOI:
关键词:
摘要: Services that share a storage system should realize the same efficiency, within their of time, as when they have to themselves. The Argon server explicitly manages its resources bound inefficiency arising from inter-service disk and cache interference in traditional systems. goal is provide each service with at least configured fraction (e.g., 0.9) throughput it achieves has itself, server--a allocated 1/nth get nearly (or more) would alone. uses automaticallyconfigured prefetch/write-back sizes insulate streaming efficiency seeks introduced by competing workloads. It explicit time quanta do for non-streaming workloads internal locality. partitions among services, based on observed access patterns, hit rate patterns others. Experiments show that, combined, these mechanisms Argon's automatic configuration achieve insulation goal.