Omer Zaki and Brian Forney
University of Wisconsin-Madison
Abstract
Distributed storage systems increase performance of storage subsystems yet introduce performance variability which can create bottlenecks in a storage system. This paper introduces performance fault masking for network-attached storage devices to decrease performance variations in distributed storage systems. Performance fault masking key components are a global cache and a cost-aware caching policy. Trace driven simulations of synthetic workloads has shown some promise with this technique. Performance variations are reduced while average request times increase for most workloads. |
Paper available as: PDF
Software available as: tar'ed and gzip'ed file