A task-specific approach to dynamic device power management for embedded system

One of the major challenges of dynamic device power management lies in the uncertain length of the idle periods of devices. In this paper, we focus on the problem of the accuracy of prediction when tasks change their request modes, leading the change of the device idle periods. We notice that there are some reasons causing the change of the idle periods. In order to capture these reasons, we first establish a two-level power management mechanism in the operating system. Then we provide a task-specific approach, which is based on the relationship between tasks and devices and takes dynamic weight of predictive value and actual value. Experiments show that our approach could save more than 40% power consumption and was more efficient and accurate than previous predictive policies.