The SCADS director: scaling a distributed storage system under stringent performance requirements

Elasticity of cloud computing environments provides an economic incentive for automatic resource allocation of stateful systems running in the cloud. However, these systems have to meet strict performance Service-Level Objectives (SLOs) expressed using upper percentiles of request latency, such as the 99th. Such latency measure-ments are very noisy, which complicates the design of the dynamic resource allocation. We design and evaluate the SCADS Director, a control framework that reconfig-ures the storage system on-the-fly in response to work-load changes using a performance model of the system. We demonstrate that such a framework can respond to both unexpected data hotspots and diurnal workload pat-terns without violating strict performance SLOs. 1