Bibliography Details

H. Chang, S. Jamin, and W. Willinger, "Internet Connectivity at the AS-level: An Optimization-Driven Modeling Approach", in ACM SIGCOMM Workshop on MoMeTools, August 2003.

Internet Connectivity at the AS-level: An Optimization-Driven Modeling Approach
Authors:	H. Chang S. Jamin W. Willinger
Published:	ACM SIGCOMM Workshop on MoMeTools, 2003
URL:	http://citeseer.ist.psu.edu/577612.html
Entry Date:	2004-06-30
Abstract:	Two ASs are connected in the Internet AS graph only if they have a business "peering relationship." By focusing on the AS subgraph AS-PC whose links represent provider-customer relationships, we develop a new optimization-driven model for Internet growth at the AS-PC 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.
Datasets:	Oregon RouteViews (late May 2001) Additional information from BGP routing tables, Looking Glass, the Routing Registry database, adding 40% more links (late May 2001)
Experiments:	AS-PC subraph of Internet derived from BGP routing information using different relationship inferrence techniques by Gao and Subramanian et. al. Successive refinements of Optimization-Driven model run for 10,000 nodes. For each iteration, one node is added with one PoP. Each existing node has a probability of gaining a new Point of Presence (PoP). First model: new node chooses a provider based on the closest Point of Presence (PoP). Second model: based on some number of metrics with uniform random values, new node makes a local Pareto optimal choice amongst providers within a certain distance (no other provider among the choices is better in all metrics). Third model: nodes are regularly randomly removed, causing customers to choose a new provider.
Results:	With more realistic assumptions, other model does not match node degree of Internet. Model is robust to changes in parameters and AS/node degree matches that of inferred Internet AS-PC subgraph. Change of provider makes node age distribution more like Internet's.
References:	Similar approach advocated and outlined in: D. Alderson, J. Doyle, R. Govindan, and W. Willinger. Toward an Optimization-Driven Framework for Designing and Generating Realistic Internet Topologies. In Proc. of HotNets-I, 2002. Compares against model of: A. Fabrikant, E. Koutsoupias, and C. H. Papadimitriou. Heuristically Optimized Trade-offs: A New Paradigm for Power Laws in the Internet. In Proc. of ICALP, 2002.