Dynamically Split the Traffic in Software Defined Network Based on Deep Reinforcement Learning

Traffic engineering (TE) can balance the traffic in the network to reduce network congestion and improve network resource utilization. The emergence of Software Defined Network (SDN) provides a more flexible and effective way to control traffic in the network. Existing TE solutions mainly focus on routing traffic via the shortest path or evenly distributing the traffic among multiple available paths, but these methods are not flexible since this static mapping of traffic to paths does not consider either the current network utilization or traffic load. Heuristics-based TE methods depends on operators' understanding of the workload and environment. Designing and implementing those methods thus take at least weeks. Furthermore, it usually takes minutes to output the solution. Inspired by recent successes in applying Deep Reinforcement Learning (DRL) techniques to solve complex control problems, we leverage DRL to control traffic in SDN. We start by building a framework which integrates the DRL algorithm into SDN. Based on this framework, we propose a modified DRL algorithm to control the traffic split ratio to multiple paths. Simulation results show that the proposed approach performs better than three baseline methods when the traffic load is dynamically changing.