Telemeter: Network Distance Prediction Base on Network Topology

Coordinates-based distance prediction algorithms can improve the performance of many Internet applications, especially in the peer-to-peer architecture and overlay construction. But some recent researches indicate that the performance gain via distance prediction based on coordinates, rather disappointed, can be significantly worse than the improvement obtained by direct measurement algorithm, and the inaccuracy of coordinates-based distance prediction would degrade the performance of applications evidently. In this paper, we propose a novel distance prediction algorithm base on network topology information, Telemeter, which uses the topology and routing information from CERNET2 's ISP, instead of coordinates mechanism. Telemeter can predict the distance accurately between any pair of nodes from CERNET2 without any end-to-end measurement overhead. We evaluate the performance of Telemeter from P2P-related application's perspective systematically, and compare it with three representative coordinates-based distance prediction algorithms (Triangulated Heuristic, GNP, and IDES). We demonstrate that Telemeter overcomes some drawback shared among coordinates-based distance prediction algorithms, and bridge the significant performance gap between measurement-based algorithm and these algorithms.