Reinforcement Learning-Based Task Scheduling Using DVFS Techniques in Mobile Devices

Mobile devices are equipped with a multi-core heterogeneous processing architecture to balance the trade-off between performance and power efficiency. To take advantage of this architecture, in this paper, we propose a task scheduling algorithm that allocates the cores to computation tasks and adjusts the cores’ operating frequency by using the dynamic voltage and frequency scaling (DVFS) technique. Our objective is to minimize the battery energy consumption of devices. We also consider offloading the tasks to a cloud server to address their ever increasingly computation demand. In order to design the task scheduling algorithm, we first formulate a combinatorial optimization problem. To efficiently solve this problem in highly dynamic situation, we then propose a reinforcement learning-based algorithm. The algorithm learns the best policy of decision making for mobile devices under dynamic situations. Through numerical studies, we investigate the performance of our proposed task scheduling algorithm for different computation task sets. Our results show that the mobile device can greatly save energy by intelligently allocating the cores and scheduling the tasks. Furthermore, we show that our proposed algorithm achieves a 6.5% reduction in energy consumption compared to an existing algorithm in the literature. These findings indicate that our proposed algorithm has great potential in enhancing the energy efficiency of mobile devices.