Detecting and locating configuration errors in IP VPNs with Graph Neural Networks

Configuration verification of networks, particularly virtual private networks, is a complex task that is required before each update on a production environment, so that network providers can ensure network availability for their customers. In this paper, we propose a Graph Neural Network (GNN) based approach to detect and locate configuration errors in IP Virtual Private Networks (VPNs). Two GNN models are proposed, one that targets routing configuration errors between customer and provider edge routers, the other targets VPN routing errors between the different provider edge routers. The objective is to provide a tool that simplifies the process of verifying end-to-end VPN configurations. We trained both models with balanced datasets containing labeled configurations generated from examples based on IMS Networks’ deployed VPNs. Results show a high level of accuracy for different sizes of VPNs (3 to 40 sites) and for two types of architectures (full-mesh and hub-and-spoke).