An Approach To System-Level Advanced Dynamic Power Management

With the rapid development of embedded system technology, the functionality of embedded system has become more and more complex. Most embedded system, however, are mobile device powered by battery. Nevertheless, the development of battery technology is at a relatively slower pace which makes most designers of embedded systems face a question: how to meet the performance and functionality requirement with the limitation of power consumption? So, power management becomes a key issue of designing embedded system.Power management can be applied in several different layers like hardware, firmware and software. Dynamic power management (DPM) discussed in this paper belongs to the software layer which is mostly implemented in OS kernel. As an efficient system-level power management policy, DPM will selectively turn down the device to its power saving state when find it idle and wake it up at suitable time to serve the next request. Its target is to reduce power consumption with the limitation of meeting system functionality.This paper first introduced several DPM policies such as Greedy, Time-out, Predictive and Stochastic Model. Along with that, this paper presented a kind of advanced dynamic power management (ADPM) policy whose main idea is to use segmented slide window for idle predicting; supporting device with multi-power saving state; using self-learning pre-wakeup algorithm; dealing with special case like too long or too short idle. In addition, this paper implemented ADPM along with some other DPM policies with software simulator and then made an experiment to compare the performance of these policies in terms of system response and power saving. The result shows that ADPM made an average improvement of about 13% in system response and 16% in power saving. In the end, this paper analysed advantages and disadvantages of ADPM over the other DPM policies according to the detail experiment data, based on which, future design will be presented.