An SDN routing algorithm based on deep reinforcement learning

In this paper, a deep reinforcement learning routing(DRL-Routing) algorithm was proposed to solve the traffic engineering(TE)problem in software defined networking(SDN). The algorithm proposed made use of more comprehensive network information to represent the state, and adopted one-to-many network configuration for routing selection. Besides, the reward function was able to adjust the network traffic of the round-trip path. The simulation results showed that DRL-Routing could obtain higher rewards. After proper training, the agent could learn a more excellent routing strategy between the switches, which increased network traffic and reduced network delay and data packet loss rate