Workload decomposition for QoS in hosted storage services

Abstract

The growing popularity of hosted storage services and shared storage infrastructure in data centers is driving the recent interest in performance isolation and QoS in storage systems. Due to the bursty nature of storage workloads, meeting the traditional response-time Service Level Agreements requires significant over provisioning of the server capacity. We present a graduated, distribution-based QoS specification for storage servers that provides cost benefits over traditional QoS models. Our method RTT partitions the workload to minimize the capacity required to meet response time requirements of any specified fraction of the requests.