On the estimation of link delay distributions by cumulant-based moment matching

The statistical distribution of packet delays across individual nodes or links embeds important information for network management. For instance, upward shift of the delay distribution could point to local congestion, node mis-functioning, or other undesirable events. Since directly measuring delays at individual links is not always possible, a problem that has been studied under the term of network delay tomography is to infer "internal" delays (per link) from "external" measurements (per path). We propose a novel estimation idea based on a combination of 2 tools from formal statistics, namely the cumulant method and the moment matching method. An outstanding feature of this approach is the extremely low implementation cost in terms of local computation and communication overhead. We formalize the estimation framework and develop a method with final closed-form solutions. The correctness of the analysis framework is validated on synthetic data obtained via simulations.