Minimizing energy consumption while satisfying the user-specified performance requirement is a primary design objective for mobile devices. To achieve this, off-the-shelf mobile devices are usually equipped with dynamic voltage and frequency scaling (DVFS)-enabled processors to trade off performance for energy reduction through online adjustment of operating points of cores. Recently, reinforcement learning algorithms, especially Q-learning, have been widely studied and show great potential for runtime DVFS control because of their adaptability to the changing environment. However, the rapid evolution of hardware and the ever-growing diversity of mobile applications increase the system complexity dramatically and make it hard for the learning agent to quickly obtain an efficient power m...[ Read more ]

Minimizing energy consumption while satisfying the user-specified performance requirement is a primary design objective for mobile devices. To achieve this, off-the-shelf mobile devices are usually equipped with dynamic voltage and frequency scaling (DVFS)-enabled processors to trade off performance for energy reduction through online adjustment of operating points of cores. Recently, reinforcement learning algorithms, especially Q-learning, have been widely studied and show great potential for runtime DVFS control because of their adaptability to the changing environment. However, the rapid evolution of hardware and the ever-growing diversity of mobile applications increase the system complexity dramatically and make it hard for the learning agent to quickly obtain an efficient power management policy. In this thesis, we propose a collaborative Q-learning-based approach to solve the DVFS control problem of mobile multicore processors. Multiple devices with different runtime conditions can acquire complementary knowledge during the learning process. Efficient knowledge sharing among devices can increase the convergence rate of the learning algorithm and potentially improve the quality of the derived power management policy. On each device, the four-phase action selection strategy is applied for efficient exploration of the action space defined by various power settings. We have also studied the impact of the temporal and spatial control granularity on the DVFS efficiency. The proposed approach is evaluated on multicore processors at different scales. Experimental results on various realistic applications show that the proposed collaborative Q-learning-based approach can achieve more energy reduction than existing methods, while providing significant speedup over the state-of-the-art individual learning algorithm.