Deep Reinforcement Learning for Multimedia Traffic Control in Software Defined Networking

Software Defined Networking (SDN) is a promising paradigm to provide centralized traffic control. Multimedia traffic control based on SDN is crucial but challenging for Quality of Experience (QoE) optimization. It is very difficult to model and control multimedia traffic because solutions mainly depend on an understanding of the network environment, which is complicated and dynamic. Inspired by the recent advances in artificial intelligence (AI) technologies, we study the adaptive multimedia traffic control mechanism leveraging Deep Reinforcement Learning (DRL). This paradigm combines deep learning with reinforcement learning, which learns solely from rewards by trial-and-error. Results demonstrate that the proposed mechanism is able to control multimedia traffic directly from experience without referring to a mathematical model.