Practical Foreground Traffic Performance Modeling for Wide Area Network

The quality of service (QoS) in wide area networks (WANs) is crucial for ensuring efficient and reliable communication. Traditional QoS evaluation approaches, such as analysis, simulation, and emulation, have shown limitations in terms of accuracy or computational efficiency. The emergence of the digital twin network paradigm has inspired data-driven schemes that aim to construct accurate and lightweight performance evaluation models. However, practical application of these schemes often encounters challenges due to their heavy dependence on complex and detailed network state information, requiring extensive measurements and reporting overhead. In this paper, we propose a practical data-driven methodology, DT-WAN, that harnesses the power of graph neural networks (GNNs) to model and evaluate QoS in WANs. DT-WAN addresses the challenge of reducing measurement complexity by focusing on foreground traffic QoS, while considering the impact of background flows with minimal information collection. Through simulation experiments, we demonstrate that DT-WAN achieves accurate prediction of QoS distributions for foreground traffic, with low Wasserstein distance.