Deep-Reinforcement-Learning-Based Joint Optimization of Task Migration and Resource Allocation for Mobile-Edge Computing

Task migration and resource allocation are essential to integrate available resources for improving the efficiency of mobile-edge computing to support various computation-intensive and delay-sensitive Internet of Things applications, which cooperative optimization has yet been well addressed to achieve desirable performance. In this article, a deep reinforcement learning (DRL)-based adaptive cooperative optimization strategy is presented to fill this gap. Task migration and resource allocation are jointly employed to adaptively integrate available resources within cooperative edge nodes for processing various randomly offloaded tasks. The policy optimization to minimize the system cost and task dropout rates is formulated as Markov decision processes. A DRL algorithm that enhances deep deterministic policy gradient with a dual experience pool is proposed to jointly optimize the task migration and resource allocation in unknown stochastic application environments. Simulation experiments have been conducted to evaluate the performance of the presented strategy, and the results illustrate that it increases system rewards by 17.9%–61.5% and reduces task dropout rate by 5.2%–31.3% comparing with benchmarks.