Coalitions and dynamics in network routing games

Congestion games are an elegant model to study the effects of resource usage and routing with strategic agents, but due to their simplicity they are inadequate to realistically model many features of traffic in computer and/or road networks. In my talk I survey our recent results on extensions of congestion games towards more realistic modeling of network routing scenarios. Our results concentrate on the existence and computational complexity of exact and approximate pure-strategy Nash and strong equilibria. Whereas in some cases it is possible to provide efficient algorithms for centralized computation, for a sufficient level of generality we can establish lower bounds by proving computational hardness results. In addition, we study the more demanding goal of reaching equilibria using decentralized protocols and the duration of the resulting improvement dynamics. More fundamentally, our treatment sheds light on the tractability of coordinated behavior of players in network routing.