This work is motivated by the observation that traffic in large networks like the Internet is not controlled by a central authority but rather by a large number of selfish agents interacting in a distributed manner. Game theory predicts that selfish behaviour in such a system leads to a Nash equilibrium, but it does not, in general, explain, how such an equilibrium can be reached. We focus on this dynamic aspect.
In the first part of this survey, we develop a dynamic model of selfish, adaptive behaviour in routing networks. We show how and under which conditions such behaviour can give rise to a stable state and analyse the convergence time. Based on these results we design a distributed algorithm to compute approximate equlibria.
In the second part, we build upon the theory developed so far in order to design an online traffic engineering protocol which proceeds by adapting route weights on the time scale of seconds. We show that our protocol converges quickly and significantly improves network performance.