Performance Differentiation for Multi-Port Arrays: A Control-Theoretic Approach

Large multi-port disk arrays typically store data for multiple applications with diverse performance requirements. Each application has a different priority, representing the relative importance of the application to the business. Currently, performance differentiation and isolation for storage workloads is implemented by statically partitioning the disk array resources, which results in over-provisioning and underutilization of resources. In this paper, the authors present a feedback controller for storage workloads to provide performance differentiation among multiple applications. Their feedback controller monitors the performance of each application and dynamically allocates the array resources so that diverse performance requirements can be met without static partitioning.