Learning Based Power Management for Periodic Real-Time Tasks

With the extensive use of portable battery powered devices in the new computing era, the power consumption problem of embedded devices has received a lot of attention in order to gain long battery life for enormous prospective. In modern processors, dynamic voltage and frequency scaling (DVFS) has been commonly used for energy reduction and temperature control. But with the variety of processor configurations, a DVFS arrangement that is energy efficient for one processor might not be appropriate for others. Furthermore, the efficiency of DVFS is also affected by the CPU intensiveness of the underlying tasks of the applications. In this paper, we propose a novel reinforcement learning based algorithm that dynamically controls the voltage and frequency of real-time systems and adjusts itself with diverse environments for optimal use of energy. Experimental results show that our proposed method can save significant amount of energy compared to other approaches.