Deep Reinforcement Learning-Based Intelligent Traffic Scheduling in Software-Defined Networks

Abstract

With the continuous increase of Internet traffic, traditional network traffic scheduling methods are facing the problems of insufficient efficiency and adaptability. Software - defined networking (SDN) provides flexible control capabilities for network traffic management, and intelligent traffic scheduling algorithms, especially scheduling methods based on deep reinforcement learning (DQN), can dynamically adapt to traffic changes in different network environments. This paper proposes an intelligent traffic scheduling algorithm based on DQN. The DQN - based algorithm effectively manages and optimizes network traffic by continuously interacting with the network environment, making real - time decisions on traffic path selection and resource allocation. It conducts experimental verification in different network scenarios. By comparing with traditional static routing and load balancing algorithms, the experimental results show that the traffic scheduling algorithm based on DQN has obvious advantages in throughput, delay, packet loss rate and load balancing effect, especially in dealing with network load fluctuations, dynamic changes and burst traffic, it can provide higher robustness and adaptability. The experiment also shows that the DQN algorithm can quickly learn and adjust the traffic path in a real - time network environment, thereby effectively reducing network congestion and delay and improving the overall performance of the network. Finally, the article also discusses the optimization direction of the algorithm, including multi - path traffic scheduling, transfer learning, etc., in order to further improve the performance of the algorithm in complex network environments.