RLCS: Towards a robust and efficient mobile edge computing resource and task based on neural network

As the Internet of Things (IoT) keeps evolving, next-generation IoT (NG-IoT) scenarios empower various applications which require low-latency connections or high bandwidth support. Mobile edge computing (MEC) is then proposed to provide users with low-latency computing services. However, the massive and heteroge-neous nature of user devices and MEC servers also brings some new challenges for resource management and task offloading. Existing works have some shortcomings because of adopting a coarse-grained task model or neglecting task graph information. The booming of artificial intelligence (AI) provides us with a more robust approach to addressing these issues. In this paper, we propose an NG-IoT user task offloading and resource scheduling architecture in the MEC scenario. We formulate our problem objective as minimizing average user task completion time (TCT). To solve the problem, we propose a Reinforcement Learning based algorithm for Container Scheduling (RLCS) and cooperating with the graph convolutional network (GCN) technique. We perform RLCS training and evaluate RLCS performance in the simulated environment. Evaluation results indicate that RLCS outperforms other baselines (e.g., reinforcement learning based algorithm, heuristic algorithm) in multiple experimental settings.