Path probing is essential to maintaining an efficient overlay network topology. However, the cost of a full-scale probing is as high as ¢¤£¦¥¨§© , which is prohibitive in large-scale overlay networks. Several methods have been proposed to reduce probing overhead, although at a cost in terms of probing completeness. In this paper, an orthogonal solution is proposed that trades probing overhead for estimation accuracy in sparse networks such as the Internet. The proposed solution uses network-level path composition information (for example, as provided by a topology server) to infer path quality without full-scale probing. The inference metrics include latency, loss rate and available bandwidth. This approach is used to design several probing algorithms, which are evaluated through analysis and simulation. The results show that the proposed method can significantly reduce probing overhead while providing bounded quality estimations for all ¥£¦¥© overlay paths. The...
Chiping Tang, Philip K. McKinley