Research on Routing Optimization of SDN Network Using Reinforcement Learning Method

The development of computer networks is making it become more complex and dynamic. How to achieve efficient package-routing in the SDN (Software Design Network) has become hot research field. SARSA-Learning is a typical Reinforcement Learning algorithm. Through the on-policy exploration and learning of the network environment, it can be used to derive the optimal decision in an unknown network environment, in this way, the network data routing and forwarding can be effectively completed. This paper yields a SARSA-Learning Routing algorithm with variable greedy function (Variable $\varepsilon$-Greedy function within SARSA-Learning Routing, V-S Routing). The V-S Routing algorithm preserves the efficiency of the SARSA-Leaning framework. The purpose of V-S Routing introduces a variable factor to $\varepsilon$-Greedy function. The V-S Routing algorithm can be dynamically calculated to represent the priority of the current state in the SDN network and to solve the problem of SDN network optimal route selection, which can avoid long package waiting queue and reduce SDN network congestion and improve the link transmission speed. The Variable $\varepsilon$-Greedy function makes the algorithm more suitable to the network environment, and it also makes V-S Routing algorithm having better generalization ability. The experimental results verify the effectiveness of the algorithm.