An Adaptive Mechanism for Fair Sharing of Storage Resources

To ensure Quality of Service (QoS) for data centers, it is critical to enforce a fair share of storage resources between competing users. Interposed schedulers are one of the most practical methods for performance isolation. Most fair queuing-based proportional sharing algorithms for existing interposed scheduler are variants of counterparts designed for network routers and may result in breaking the fairness of proportional sharing required by Service Level Agreements for storage systems. This paper presents a novel algorithm to address this problem. As an extension of the fair queuing-based algorithm, it can dynamically adapt to the performance variation of storage systems and guarantee a fair sharing of resources as well as satisfying the minimal performance requirements for different clients. The design and performance evaluation are presented.