Performance of Wireless Heterogeneous Networks with Always-best-connected Users

We consider a geographic area covered by two wireless networks. Assuming delay-sensitive users, we study the loss of efficiency of the user equilibrium (the Price of Anarchy) in terms of total delay, with M/M/l delay functions on each network. The user equilibrium is proved to be less efficient when the network is very heterogeneous, i.e. the two networks have different capacities. In order to elicit coordination among users, we suggest to use marginal cost pricing. We investigate the computation of the optimal taxes to use, and give several arguments in favor of the technical feasibility of such a scheme. Applying taxes therefore seems particularly well-suited to improve the overall performance of a network selection game with selfish users.