A Deep Reinforcement Learning Approach for Software-Defined Wide Area Network Routing Optimization

The paper discusses the shortcomings found in the conventional SD-WAN routing, which is based on predefined rules and heuristics, resulting in the inefficiency of dynamic shifts of the networks. It suggests that the introduction of DRL, especially Q-learning and other RL algorithms, would greatly add to the capacity of SD-WAN to dynamically change depending on the variation of traffic, network conditions, and other factors. Such a method enables the maximization of the Quality of Service (QoS) parameters such as latency, jitter, throughput, and packet loss, which enhances network efficiency and decreases network outages. The article highlights the possibility of AI-led SD-WAN in the automation of the decision process, providing cost-efficient solutions and high carrying capacities of the network, at least concerning the low-latency services, such as video calls and voice-over-IP. The paper ends with emphasis on the success of DRL over conventional routing protocols as well as other machine learning solutions, which makes it a revolutionary solution for the future of SD-WAN technologies.