Energy-Efficient Scheduling and Optimization in Edge Computing based on Deep Q-Networks

Mobile Edge Computing (MEC) became a promising paradigm that provides computational and storage capabilities for enhancing user devices that run heavy mobile applications. MEC brings computational resources closer to the network's edge, reducing latency and bandwidth consumption by processing data without transporting computational tasks to distant cloud servers. Reduced latency, guaranteed by MEC because of the geographical proximity, allows improved quality of service (QoS). Although MEC solves the problem of limited computational capabilities of user devices, this computing paradigm requires a strategic approach to manage power consumption, especially in contemporary MEC environments that are characterized by dynamic client workloads on the system, increasing the complexity of the problem. In order to deal with this issue, we proposed a Deep Q-Network (DQN)-based offloading technique to reduce power consumption with such dynamic workloads considerably. This work also highlighted the potential of reinforcement learning in dynamic and heterogeneous MEC environments.