Skip to Main Content
Storage consolidation as a perspective paradigm inevitably leads to the extensive installations of shared storage servers in product environments. However, owing to the dynamics of both workloads and storage systems, it is pragmatic only if each workload accessing common storage servers can surely possess a specified minimum share of system resources even when competing with other workloads and consequently obtain predictable quality of service (QoS). This paper presents an I/O scheduling framework for shared storage servers. The eMuse algorithm in the framework employs a dynamic assignment mechanism that not only accommodates a weighted bandwidth share for every active workload, but also fulfills their latency requirements through a fair queuing policy. Experimental results demonstrate that our scheduling framework can accomplish performance isolation among multiple competing workloads as well as the effective utilization of system resources.