The effects of selfish pricing and collusion in network routing games

This thesis is devoted to the study of traffic routing in large-scale heterogeneously created and operated networks. In many such networks it is often infeasible to directly enforce optimal traffic routes. Instead, the network users are left to choose the paths that best match their objectives, and they often do so in an uncoordinated and purely self-interested fashion. Such selfish behavior usually results in a socially inefficient outcome, whereby the users overcrowd the better paths in the network and thus reduce the overall quality of the routing. An important question in this setting is to estimate the extent of this inefficiency. We address this question in two routing contexts. In the first context, the underlying network is owned by service providers, who compete for traffic by setting prices for their links. The network users choose their routes by trading off the delay on the route against the price being charged for using that route. The service providers extract profit proportional to the price they charge from the volume of traffic that uses their link, and they are solely interested in maximizing that profit. Our goal in this setting is to study the extent of social inefficiency that is caused by the competitive pricing on the part of the service providers. The second context involves the existence of coalitions among the network users. The coalitions are assumed to display full collaboration in that each coalition chooses routes for its members with the objective of minimizing the average delay within the coalition. While coalitions clearly have the potential to improve the overall quality of the routing (in particular if all users form a single coalition, the routing will be optimal), we demonstrate that in many routing models coalitions can also harm the quality of the routing. In light of this, we study the extent by which forming coalitions can decrease the social welfare of the outcomes achieved in various routing scenarios.