Leveraging Graph Attention Network for SDN Routing Performance Prediction

Abstract

With the continuous expansion of the network scale, the network topology is becoming increasingly complex, and the link status between network devices will also change dynamically according to real-time load, link quality and other factors, increasing the difficulty of solving network optimization problems. Therefore, an efficient and fine-grained network model is essential to achieve the goal of efficient and autonomous network optimization. As the development direction of future network architecture, Software Defined Networking (SDN) technology can effectively set up routing schemes and flexibly control network traffic by separating data plane and control plane. In the process of routing scheme optimization, the key is to accurately estimate the network performance under a given routing scheme. In this paper, we propose a SDN routing performance prediction model based on graph attention network. Based on graph attention network, this model models the relationship between physical links and routing scheme paths in the network. Under the given routing scheme and network traffic, it accurately estimates each end-to-end performance index in the network to assist in optimizing the routing scheme.