Selfish peering and routing in the Internet

Abstract

The Internet is a loose amalgamation of independent service providers acting in their own self-interest. We examine the implications of this economic reality on peering relationships. Specifically, we consider how the incentives of the providers might determine where they choose to interconnect with each other. We consider a game where two selfish network providers must establish peering points between their respective network graphs, given knowledge of traffic conditions and a nearest-exit routing policy for out-going traffic, as well as costs based on congestion and peering connectivity. We focus on the pairwise stability equilibrium concept and use a stochastic procedure to solve for the stochastically pairwise stable configurations. Stochastically stable networks are selected for their robustness to deviations in strategy and are therefore posited as the more likely networks to emerge in a dynamic setting. We note a paucity of stochastically stable peering configurations under asymmetric conditions, particularly to unequal interdomain traffic flow, with adverse effects on system-wide efficiency. Under bilateral flow conditions, we find that as the cost associated with the establishment of peering links approaches zero, the variance in the number of peering links of stochastically pairwise stable equilibria increases dramatically.