GNN-Based SDN Admission Control in Beyond 5G Networks.

This paper proposes a novel approach to Software-Defined Networking (SDN) Admission Control (AC) based on Graph Neural Networks (GNNs) for Beyond 5G (B5G). AC is a critical function in SDN, as it determines which traffic flow to pass by the network and which should be rejected. GNNs are a type of Neural Networks (NNs) that are able to learn how to make real-time AC decisions by training on pre-existing data, including network topologies and traffic characteristics. The solution we propose is made of two layers: (i) Network Delay Predictor (NetDelP) leveraging on the RouteNet-Fermi GNN model, used to predict the network latency for different topologies and traffic patterns. (ii) Admission Control Agent (AdConAgt) supporting the SDN and used to regulate the traffic flow in the network. The outlined concept is able to manage large-scale and complex topology networks with optimized Key Performance Indicators (KPIs) such as network latency and Packet Loss Rate (PLR). The envisioned approach is evaluated with various network topology scales and classes of traffic. The obtained results outperform the SDN-Shortest Path (SDN-SP) solution by demonstrating the ability of our proposal to guarantee the End-To-End (E2E) latency and prevent link congestion in order to meet the QoS requirements.