Graph Neural Network for Real-Time Simulation of SDN-Enabled Communication

Efficient and reliable communication networks are essential for naval operations and broader wire or wireless communications. Software-Defined Networking (SDN) has emerged as a pivotal tool in the management of communication networks to achieve this goal efficiently. A key advantage of SDN is its ability to configure networks to reduce packet delay, jitter, and loss. To effectively manage these metrics, predictive network behavior is necessary. While traditional network simulators like OMNeT++ serve this purpose, they become computationally intensive with large network topologies. Addressing this limitation, recent research has shifted towards employing deep learning architectures for network behavior prediction. Graphical Neural Networks (GNN), with their structural resemblance to network topologies and graph data structures, are particularly suited for this application. One notable GNN architecture, RouteNet, has demonstrated proficiency in predicting network delay and jitter, rivaling traditional network simulators in accuracy. This paper studies how RouteNet can be used to evaluate network behavior using traffic statistics. The results of the simulation test have validated the effectiveness of RouteNet in predicting the performance of the SDN network through the Mininet environment.