An Intelligent Routing Approach for SDN Based on Improved DuelingDQN Algorithm

Existing intelligent routing methods make routing decisions based on limited link information in data center networks with surging traffic, making it difficult to adapt to complex network environments and dynamic traffic loads that cause congestion and low bandwidth utilization. In order to improve the traffic transmission efficiency and network reliability, this paper proposes a smart routing algorithm (DRL-PD, Deep Reinforcement Learning-Prioritization and Dijkstra) based on improved DuelingDQN. The algorithm combines the Software Defined Networking (SDN) controller to monitor the network link information and maps it into the state space of the intelligences for training decisions, designs the intelligent routing model as a multidimensional reward weight, proposes a Dijkstra a priori exploration strategy based on Epsilon’s greed to intelligently adjust the routing and forwarding strategy, thus adapting to the changes in the network traffic, alleviating link congestion, and improving the traffic transmission efficiency and performance of the network. The experimental results show that DRL-PD has the effectiveness and stability of fast convergence on training compared to Deep Deterministic Policy Gradient (DDPG) and Proximal Policy Optimization (PPO) algorithms. In terms of routing effectiveness, DRL-PD has a good performance in improving network throughput and reducing data packet loss under guaranteed transmission tasks compared to Deep Reinforcement Learning-Traffic Engineering (DRL-TE) method, state-action-reward-state-action (SARSA) and Dijkstra routing algorithm.