Adaptive control for power management

Abstract

Power management techniques are applied to several types of computer systems, such as desktops, servers and embedded systems. For all of those systems, the use of a power management technique is expected to decrease power consumption without degrading performance. Therefore, a variety of power management techniques tailored for specific hardware architectures and the power management mechanisms for computer systems have been investigated. Feedback control constitutes a very promising technique for power management, because it provides some theoretical guarantees in terms of tracking accuracy and system stability. However, a computer system is subjected to unknown and nonstationary workloads. Therefore a feedback control based power management policy must be capable of adapting itself to the fluctuations of the workload, and still guarantee optimal performance with minimum power consumption. This paper proposes an adaptive control technique for power management considering such requirements. The experimental results have shown that the proposed adaptive control technique outperformed other stateof-the art policies both in power and system performance.