Deep Reinforcement Learning Based Task Scheduling in Edge Computing Networks

With the rapid development of 5G mobile networks services, massive data explodes in the network edge. Cloud computing services suffer from long latency and huge bandwidth requirement. Edge computing has become the key technology of reducing service delay and traffic load in 5G mobile networks. However, how to intelligently schedule tasks in the edge computing environment is still a critical challenge. In this paper, we define the optimization problem of minimizing the delay for task scheduling in the cloud-edge network architecture. The problem is proved NP-hard and modeled following Markov decision process. We design a cloud-edge collaboration scheduling algorithm based on asynchronous advantage actor-critic (CECS-A3C). Simulation results show that the proposed algorithm has good convergence speed and can reduce the task processing time by an average of 28.3% and 46.1% compared with the existing DQN and RL-G algorithms, while keeping the performance scalable.