TR-clustering: Alleviating the impact of false clustering on P2P overlay networks

• Published in: Computer networks (Amsterdam, Netherlands : 1999) Vol. 52; no. 17; pp. 3185 - 3204
• Main Authors: Sànchez-Artigas, Marc, García-López, Pedro, Gómez-Skarmeta, Antonio F., Santa, José
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier B.V 08.12.2008; Elsevier; Elsevier Sequoia S.A
• Subjects: Applied sciences; Exact sciences and technology; Interconnected networks; Internet; Internet Coordinate Systems; Internet Protocol; Miscellaneous; Network Clustering; Network topologies; Networks and services in france and abroad; Organization and planning of networks (techniques and equipments); Peer to peer computing; Peer-to-Peer Systems; Routers; Simulation; Studies; Systems, networks and services of telecommunications; Telecommunications; Telecommunications and information theory; Teleprocessing networks. Isdn; Transmission and modulation (techniques and equipments); Network Clustering; Peer-to-Peer Systems; Internet Coordinate Systems; Performance evaluation; Automatic classification; Peer to peer; Similarity; Network architecture; Shortest path; Router; Resource sharing; Coordinate system; Signal classification; Data broadcast; Information dissemination; Multicast; Topological structure; Overlay network; Computer network; Simulation; Telecommunication network; Internet
• ISSN: 1389-1286; 1872-7069; 1872-7069
• DOI: 10.1016/j.comnet.2008.08.011

Designing topologically-aware overlays is a recurrent subject in peer-to-peer research. Although there exists a plethora of approaches, Internet coordinate systems such as GNP (which attempt to predict the pair-wise O( N 2) latencies between N nodes using only O( N) measurements) have become the most attractive approach to make the overlay connectivity structures congruent with the underlying IP-level network topology. With appropriate input, coordinate systems allow complex distributed problems to be solved geometrically, including multicast, server selection, etc. For these applications, and presumably others like that, exact topological information is not required and it is sufficient to use informative hints about the relative positions of Internet clients. Clustering operation, which attempts to partition a set of objects into several subsets that are distinguishable under some criterion of similarity, could significantly ease these operations. However, when the main objective is clustering nodes, Internet coordinate systems present strong limitations to identify the right clusters, a problem known as false clustering. In this work, the authors answer a fundamental question that has been obscured in proximity techniques so far: how often false clustering happens in reality and how much this affects the overall performance of an overlay. To that effect, the authors present a novel approach called TR-Clustering to cluster nodes in overlay networks based on their physical positions on the Internet. To be specific, TR-Clustering uses the Internet routers with high vertex betweenness centrality to cluster participating nodes. Informally, the betweenness centrality of a router is defined as the fraction of shortest paths between all pairs of nodes running through it. Simulation results illustrate that TR-Clustering is superior to existing techniques, with less than a 5% of falsely clustered peers (of course, relative to the datasets utilized in their evaluation).