RouteNet-Gauss is a novel network simulation methodology that integrates a testbed network and a machine learning (ML) model to overcome the computational cost and accuracy limitations of conventional discrete event simulation (DES) methods. Leveraging the testbed as a hardware accelerator, it rapidly generates a high-quality training dataset and simulates network scenarios that closely resemble real-world environments. Experimental results demonstrate that RouteNet-Gauss reduces prediction errors by up to 95% and inference times by 488x compared to state-of-the-art DES-based methods. Its modular architecture dynamically configures the model based on characteristics such as network topology and routing, enabling understanding and generalization across diverse network configurations, including networks up to 10x the size of the training data. Furthermore, it supports Temporal Aggregate Performance Estimation (TAPE), providing configurable temporal granularity and maintaining high accuracy for flow performance metrics.
