Adaptive Dynamic Power Management for Non-Stationary Self-Similar Requests
非平稳自相似业务下自适应动态功耗管理

Dynamic power management (DPM) is a design methodology aiming at reducing power consumption of electronic systems. The effectiveness of a power management scheme depends critically on the power management policy and control algorithm. In this work, it is found that the exponential distribution supposition taken by traditional queuing theory is not suitable to DPM, and the timeout policy is enough for DMP. The reason that the simple timeout policy could reduce much energy consumption of computer devices is the self-similarity nature of computer service requests. This paper proposes an adaptive control algorithm for DPM of embedded operating system when the idle time length fits Pareto distribution. It adopts the Trimmed Mean Estimator to realize the robust efficient estimation of the tail index parameter of Pareto distribution for small sample size, and is implemented using window-size-based adaptive control method. Simulation results show that the algorithm presented in this paper is robust. The competitive ratio is reduced to 1.24 and even 1.47 when the delay is smaller than 0.10. Overhead of the adaptive control algorithm is low.