A clustering approach for exploring the Internet structure

Abstract

The paper proposes several clustering algorithms to explore the topology of the AS-graph. Using these algorithms, we are able to view the Internet topology, as encapsulated by the AS-graph, at three levels of abstraction: (1) a low-resolution view, which shows a coarse cluster cover of the whole AS-graph; (2) a mid-resolution view, showing the relationships between dense cores inside the coarse clusters; (3) a high-resolution view of individual high-density cores. Using these tools we discover that the Internet's central dense core is at least twice as large as what was previously reported. Attached to the central dense core, in what is known as the transit core, our clustering tools discovered several clusters, which we call the regional dense cores. Interestingly, these clusters are as dense as the central core itself. Our tools highlight the fact that the regional dense cores are mostly connected to the central dense core and less well connected among themselves.