An online predictive control framework for designing self-managing computing systems

Abstract

This paper describes a control-theoretic framework to design self-managing computing systems that continually optimize their quality-of-service (QoS) in response to changing computing demands and environmental conditions. A predictive control technique is used to tune the performance of individual system components based on their forecast behavior over a look-ahead horizon. As a specific application of this control technique, the power consumed by a single computer processing a time-varying workload is minimized. The proposed method can also be extended to a distributed setting where a global controller manages inter-component interactions such that system-wide QoS requirements are satisfied.