Optimizing Load Balancing Routing Mechanisms with Evolutionary Computation

Link State routing protocols, such as Open Shortest Path First (OSPF), are widely applied to intra-domain routing in todays IP networks. They provide a good scalability without lost of simplicity. A router running OSPF distributes traffic uniformly over Equal-cost Multi-path (ECMP), enabling a better distribution of packets among the existent links. More recently, other load balancing strategies, that consider non even splitting of traffic, have been put forward. Such is the case of the Distributed Exponentially-weighted Flow SpliTting (DEFT), that enables traffic to be directed through non equal-cost multi-paths, while preserving the OSPF simplicity. As the optimal link weight computation is known to be NP-hard, intelligence heuristics are particularly suited to address this optimization problem. In this context, this work compares the solutions provided by Evolutionary Algorithms (EA) for the weight setting problem, considering both ECMP and DEFT load balancing alternatives. In addition to a single objective network congestion optimization problem, both load balancing schemes are also applied to a multi-objective optimization approach able to attain routing configurations resilient to traffic demand variations.