Deep Reinforcement Learning-Based Traffic Engineering in Cloud-Edge-End Collaboration Environments

With the rapid growth and widespread usage of smart devices, the emerging Internet service represented by face recognition and video streaming have brought great traffic pressure for the existing mobile communication networks. Moreover, the users' mobility makes traffic engineering more complicated. Cloud-edgeend coordination has recently been regarded as effective solution to improve traffic distribution. In order to reduce redundant content transmission and improve end-users' quality of experience in mobile cloud-edge-end cooperation environments, we propose a traffic engineering algorithm based on deep reinforcement learning (DRL) in this paper to tackle these challenges. We model the optimal network traffic problem as a maximal traffic offloading model, where network devices' caching capacity is considered and the mobile users' same requests will be aggregated. We design a new DRL scheme to solve the maximal traffic offloading model based on request history and timely network status in the system. Numerical results show that the proposed policy demonstrates much better compared to the existing popular counterparts in cloud-edge-side collaboration networks.