Loss-based Inference of Multicast Network Topology 1

Inference of network-internal characteristics from the en dto-end behavior of multicast traffic has been proposed in recent papers. Probes are multicast from the source to receivers; the record of which probes reached each receiver is used to infer internal loss probabilities. Probe loss is assumed to occur independently across links and between probes. In [1, 2] the Maximum Likelihood Estimator (MLE) of link loss probabilities was determined for general logic al multicast trees. It was shown that the estimator is reasonably robust with respect to violations of this assumption, at least on the scale found in realistic network simulations . In a related paper [5], the focus was to group multicast receivers that share the same set of network bottlenecks from the source. This used the special case of the estimator for binary trees in [1] in order to estimate loss on the common portion of the path from the source to each receiver. An extension of the binary approach to treat more general trees was also proposed.