A Gradient Learning Optimization for Dynamic Power Management

Dynamic power management (DPM) is a power dissipation reduction technology aimed to adapting the power and performance of a system to its workload. In this paper, we propose a gradient learning optimization method for the DPM problem. Our method does not depend on accurate model parameters and is only based on a single sample path of system. Thus, there is no any transition probability to be calculated. Moreover, the new method only need less storage for the performance optimization. Simulation results demonstrate the applicability of the proposed method.