A Practical Network Digital Twin for IGP Weight Optimization

Current research on Network Digital Twins (NDT) focuses on building models that predict the behavior of the network on simulations rather than building a complete NDT that automatically monitors a physical network and optimizes it according to its current state and traffic flows. In this paper, we implement an online learning NDT for Interior Gateway Protocol (IGP) weight optimization that monitors a physical network to find IGP weight configurations based on a selected objective, such as minimizing average traffic delay, and automatically applies them to the network. We addressed the practical issues of implementing a NDT on a real network, such as detecting and dealing with an improperly trained digital model of the network. Our results indicate that our NDT can effectively optimize a physical testbed and that fine-tuning its digital model on high traffic scenarios improves the performance of the NDT on a small network.