From advertising profits to bandwidth prices

Abstract

The economics of peering [18], is among the thorniest debates affecting the Internet, but yet, least understood ones. The term peering refers to the interconnection between networks for the purpose of exchanging traffic directly between them. Classic unpaid peering played a crucial role in the evolution of the Internet. It was usually set up between local access ISPs of similar size for the purpose of avoiding charges and longer paths through upstream “transit” providers. This has changed recently with the establishment of peering between dissimilar networks, namely Access ISPs (AISPs) and Content and Service Providers (CSPs). In contrast with classic unpaid peering, the peering between AISPs and CSPs is primarily driven by the need to offer premium quality to emerging services such as video, search, online social networks, and gaming. Cablevision and Netflix recently signed such a premium peering agreement [13]. Classic unpaid peering [18] was justified on the basis of traffic symmetry, which no longer exists since CSPs inject into A-ISPs several orders of magnitude more traffic than they receive from them [6]. This has opened the door to fierce conflicts between A-ISPs and CSPs about who should pay whom and at what rate. Therein peering coordinators have been arguing about payments and have tried to relate them to the question of “who benefits the most from the premium peering?”. They have focused mainly on benefits from reduced transit costs [18], without handling the question of “who can monetize better the superior traffic delivery quality?”. Many of the proposed models miss important details of the conflict, as prices are derived based on a bilateral basis rather than in a competitive market. Moreover, they are not driven by real data, and hence, cannot derive quantitative results, i.e., actual prices for premium peering, and consequently can not be used in peering negotiations. Motivated by the above, we propose a novel framework capable to analyze premium peering agreements quantitatively. Specific contributions can be broken down as follows: • Modeling: We model both costs and new profits due to improved QoE that translates into increased engagement time of existing customers, and incoming churn of new customers taken from competitors. We allocate the total surplus by solving a Nash bargaining problem which outputs fair side-payments. • Data driven approach: A methodological contribution