What is Happening from Behind? - Making the Impact of Internet Topology Visible

Abstract

The Internet topology has evolved over the past decades in an evolutionary process and continues to grow. Recently, it has attracted much attention from the networking and physics communities, as it forms a unique operational instance of a planetary-scale network environment. Several measurement projects observing the Internet have been undertaken over the past years, out of which Skitter, its successor Archipelago (Ark) and Dimes have established as continuous recordings of the vivid process of network formation. In this paper we compare Internet measurement data obtained from Skitter, Ark and Dimes by analyzing the Internet node degree distributions and correlations at IP node and router level. This comparative analysis was enabled by a data conversion and processing tool-chain implemented as an extension to the BRITE topology generator which we introduce, as well. Our results show significant differences in higher nodal degrees. Correlation analysis indicates that DIMES scans discover Internet links to a fairly uniform degree, while parts remain invisible within Skitter and Ark data. Mid-range, oscillating spatial autocorrelations are discovered as a signature of memory effects in Internet topology. Further on we analyze implications of the Internet structure as attained in both, its core and edge vicinities. Mobile multicast routing performance is quantized by the number of states minimally required for servicing listener or sender mobility. Results show a surprisingly low mobility overhead as compared to general multicast forwarding state management. As continuous mobility handovers necessarily occur between access routers located in geographic vicinity, we finally investigate the hypothesis that geographically adjacent edge networks attain a reduced network distances as compared to arbitrary Internet nodes. The evaluation of edge distance distributions in different regions for IP ranges, clustered according to their geographic location, reveals a stable correlation of geographic and network proximity at Internet edges.