Detection of shared congestion points based on likelihood ratio test

In recent years, the spreading of applications serving multiple senders and/or receivers, like teleconferencing and downloading from multiple mirror sites, has stimulated the conception of multiple-flow strategies toward a cooperative congestion control. In this framework the important issues arise of detecting if two (or more) data flows experience network congestion and, subsequently, if they share the same congestion points. In this paper we face the latter topic, namely we present a method for detecting shared bottlenecks in a multiple sender - single receiver instance by statistically processing the interarrival times of congested streams, based on the hypothesis of long-range dependent data. The proposed detection strategy is tested in typical network scenarios.