Distributed Utility Maximization From the Edge in IP Networks

To improve bandwidth sharing in IP networks, load balancing and rate control are key traffic engineering ingredients. In this context, we propose a fully distributed load balancing and rate allocation mechanism that operates only from the edge. Each access router is able to determine target rates over multiple paths for different traffic aggregates based on already available link state information and, some small and optional, information received from other edge devices. Our distributed utility maximization solution provides a feasible rate allocation at each iteration with diminishing returns. Through numerical results on a variety of instances, we show that it converges to near optimal solutions after a few iterations. Thanks to packet-level simulations on an SD-WAN scenario, we also show that it can well prioritize traffic over centralized and legacy solutions.