Deep Q-learning based Resource Scheduling in IoT Edge Computing

Edge computing (EC) has emerged as a viable solution for resource-intensive Internet of Things (IoT) applications seeking low-latency services at the network edge. However, the limited computing power of edge servers poses challenges in scheduling application jobs and allocating resources effectively. This study addresses the task scheduling and resource allocation problem in the EC scenario, aiming to maximize the long-term satisfaction of tasks assigned to virtual machines (VMs) deployed at the edge server. Formulated as a Markov decision process (MDP), the problem involves states, actions, state transitions, and rewards. Subsequently, a Deep Q Neural Network (DQN)-based resource allocation algorithm is developed to devise an optimal offloading decision and resource allocation scheme in an edge computing environment with constrained resources and latency limits. Simulation results demonstrate that the proposed DQN-based methods outperform existing approaches in the literature, achieving superior performance in terms of average waiting time and efficiency. This research contributes to enhancing the effectiveness and efficiency of resource allocation in edge computing environments, thereby facilitating the deployment of latency-sensitive IoT applications.