Internet connectivity at the AS-level: an optimization-driven modeling approach

Abstract

Two ASs are connected in the Internet AS graph only if they have a business "peering relationship." By focusing on the AS subgraph ASPC whose links represent provider-customer relationships, we develop a new optimization-driven model for Internet growth at the ASPC level. The model's defining feature is an explicit construction of a novel class of intuitive, multi-objective, local optimizations by which the different customer ASs determine in a fully distributed and decentralized fashion their "best" upstream provider AS. Key criteria that are explicitly accounted for in the formulation of these multi-objective optimization problems are (i) AS-geography, i.e., locality and number of PoPs within individual ASs; (ii) AS-specific business models, abstract toy models that describe how individual ASs choose their "best" provider; and (iii) AS evolution, a historic account of the "lives" of individual ASs in a dynamic ISP market. We show that the resulting model is broadly robust, perforce yields graphs that match inferred AS connectivity with respect to a number of different metrics, and is ideal for exploring the impact of new peering incentives or policies on AS-level connectivity.