Evolutionary routing-path selection in congested communication networks

This paper proposes an evolutionary approach to the network traffic optimization under the constraint of congestion avoidance. The individuals of the evolving population directly represents a set of paths in a network, and corresponding crossover and mutation operators are provided. The optimization is a global one, i.e. it will not optimize the paths independently but also taking link sharing into account. To avoid the situation that the optimization will result in no traffic for some of the senders (which is also an element of the feasible space in congestion avoidance), we use the user fairness concept. A general approach to user fairness is also provided. The fitness of an individual (path set) is computed from the total traffic in the maxmin fairness state. Experiments on certain graph structures were performed. The results were compared with a path selection strategy based on single path evaluation only. The experiments for networks in the dimension 10 to 80 nodes demonstrate that an increase in performance of around 10% can be achieved in many cases, even with rather small population sizes and number of generations.