Low power scheduling based on statistics

Dynamic Voltage Scaling(DVS) and Dynamic Power Management(DPM) are two major techniques for reducing energy consumption in computer systems today. In this paper, we will propose a model that can capture the key characteristics of power management in today's advance computer systems. In this model, a processor in active mode can run at a number of speeds and each has a different energy consumption rate. The processor can also sleep in a number of sleep modes and each has a different wake-up latency and energy consumption rate. We study the power-driven task scheduling problem in this model and devise an effective method to determine the speed and mode of operation at each time step in order to finish a set of input tasks successfully with an objective to minimize the total power consumption. We found that the offline version of this scheduling problem can be formulated as a quadratic program, and can be solved optimally in practice. We will also present an effective online algorithm to solve this problem that can achieve good competitive ratios in terms of power consumption in comparison with the optimal offline results. We compare our algorithm with other two state-of-the-art techniques. Our experimental results reduce on average 66% and 9% energy respectively.