Min Li, Xiaobo Wu, Menglian Zhao
Dec 9, 2004
Journal name not available for this finding
Dynamic Power Management (DPM) is an effective power reduction technique to dynamically control power state of system components. Although there are already a lot of papers on DPM, few of them present robust and adaptive solution to handle the inherent time varying behavior of real world system. In this paper, we propose techniques for targeting DPM on the time varying behaviors. In our approach, we apply efficient policy optimization method to generate online power management policy to achieve adaptiveness. Moreover, in order to handle the small scale varying behavior (perturbation) that tends to significantly degrade the performance of DPM, we apply the Bounded Parameter Markov Decision Process and Interval Value Iteration to derive robust policy tolerant to perturbations. Simulation results show that the proposed techniques are effective.