CTGANs for Generating Synthetic Traffic Matrices in Wide Area Networks

This article is an overview of how to use Generative Adversarial Networks (GANs) for generating synthetic traffic matrices for network research. We show Conditional Tabular Generative Adversarial Networks (CTGANs), a method based on GANs that accurately captures the dynamics of 167 real traffic matrices from the US Abilene topology. This method scores more than 80% in quality score and preserves the statistical properties of the original dataset. Also, when evaluating the traffic matrices on the topology, the synthetic data accurately mimics the original data, showing average link load values similar to the original ones. The code used in the experiments is also provided to the community in GitHub for further use and experimentation.