An Intelligent Routing Mechanism Based on Deep Reinforcement Learning

With the continuous development of network technologies such as cloud computing and mobile computing, the demand of services is increasing significantly, which brings exponential grow of network traffic. In order to solve the routing problem of high concurrent traffic on the premise of ensuring QoS, an intelligent routing optimization mechanism is needed to realize the traffic efficient forwarding. Software-defined networks(SDN) provide the basis for programmable routing. Additionally, SDN and IPv6 are compatible with natural programmable capabilities, and technologies such as BIERv6 and iFIT can be deployed on it. In SDN, the controller can control the traffic routing by modifying the routing table in the switch, but the inefficient routing decision will lead to the overload of the network link and increase the end-to-end transmission delay. Deep reinforcement learning has the ability of intelligent decision-making and processing large-scale high-dimensional data, which is suitable for solving decision optimization problems such as network routing decision-making. In this paper, we proposed an intelligent routing optimization model based on SDN. The model uses deep reinforcement learning to learn the relationship between the traffic load of switches and network performance, and uses Nadam optimizer to update gradients to determine the link weights in the network. SDN controller determines the routing forwarding path through the link weights.