Request-Aware Task Offloading in Mobile Edge Computing via Deep Reinforcement Learning

The popularization of smart mobile devices has brought about the emergence of a new generation of mobile applications, such as face recognition and virtual reality. The existing mobile edge computing technology can offload tasks to the edge server for computation through the wireless channel, thereby satisfying the low delay requirement of the applications. However, due to the limited computing resources, a single-edge server cannot satisfy the offloading requirements of all users. Request Aware Task Offloading (RATO) scheme was proposed aiming at the problem that the limited edge server computing resources made it impossible to meet the requirements of task completion delay and device energy consumption with the optimization objective to minimize the weighted total overhead (including the mobile device's delay performance metric and energy consumption performance metric). Specifically, we first formulated the task offloading and resource allocation problem as a Markov Decision Process (MDP). After that, a deep reinforcement learning algorithm based on Deep Q Network was developed to solve the optimal offloading scheme. The simulation results show that the weighted total overhead of the RATODQN is lower than that of the existing schemes by 41.59% on average, thereby effectively improving the user's QoE.