Abstract
Mobile-Edge Computing (MEC) is a paradigm that performs computational operations on edge devices over the network. It improves Quality of Service (QoS) by reducing both latency and bandwidth usage. However, MEC also raises concerns about power consumption and the need to optimize energy expenditure. In this paper, we present a Deep Q-Network-based (DQN-based) scheduling framework that optimizes task offloading to reduce power consumption in MEC environments. We build a realistic environment by deploying computational nodes and simulating workloads derived from the Google Trace dataset. The dataset is used to mimic fluctuating datacenter workloads, and the tasks are homogeneous executions of Dijkstra's algorithm. The DQN algorithm is evaluated against state-of-the art baselines from recent papers using the same environment. Our findings highlight the potential of lightweight reinforcement-learning techniques for managing dynamic MEC systems.