Intelligent routing optimization for SDN based on PPO and GNN

Abstract

With the continuous increase of network scale, a variety of emerging applications (e.g., video streaming) continue to emerge, which puts forward differentiated requirements for quality of service (QoS), such as extremely low latency, high bandwidth, low packet loss rate, etc. To this effect, an intelligent routing method is needed to meet various traffic QoS requirements. However, existing routing optimization QoS schemes either lack guaranteed QoS performance or cannot generalize to invisible network topologies. To address the problem, this article innovatively proposes the QoS intelligent routing method PPO-R, which combines PPO and GNN-based routing optimization for SDN. The algorithm generates multiple disjoint candidate paths by designing a redundant tree algorithm, infers the near optimal traffic splitting ratios on the pre-selected path, and reasonably allocates traffic to different paths based on the obtained traffic splitting ratios. Compared to the baseline scheme DQS, in the medium network topology Germany-50, the QoS performance of PPO-R is about 10.6% higher than that of DQS. The experimental results show that the PPO-R algorithm can achieve better QoS routing optimization results, consume less computing resources, and significantly improve the robustness and generalization of QoS routing.