Adaptive Multipath Routing : Collaborative Load Balancing and Topological Characterization

Context Routing is one of the keys for controlling network resources. In current IP networks, routing algorithms objective is to find either a single best path between each pair of routers, or several paths with equal and optimal costs (ECMP, [1,2]). More generally, a multipath routing algorithm is an algorithm that allows finding multiple routes between two points on the network, regardless of the constraints on the costs of these paths. Such protocols have been proposed in the literature ([3,4,5,6,7]). At the router level and towards a given destination, the goal is to provide not only one but several outgoing interfaces. The hopby-hop composition of these alternatives provides a wide variety of routes which are distinct or not. There are many advantages with multipath routing. On the one hand, calculating multiple paths with anticipation allows to quickly adjusting routes and overcoming failures that may occur in the network. In a distributed environment, routers can make local routing decisions that are faster than a global routing calculation. On the other hand, different paths can be used simultaneously to take advantage of bandwidth available on each of them. The main key is then to decide the ideal distribution of traffic among these multiple routes: this is called adaptive routing. This way, two approaches can be considered. The first is the assumption that one central equipment has the complete knowledge of traffic demands: the load balancing then consists in optimizing an objective function (of overall resource usage, used bandwidth, etc.). This view has been the subject of many studies. The second approach is to assign routing decisions to different actors in the network: they are asked to collaborate together to efficiently distribute the load across the network. The study of solutions within this last approach motivates this thesis.