Estimation of network link loss rates via chaining in multicast trees

Of increasing importance is estimation of internal link parameters in communications networks. Multicast probes are a way to gather statistics about internal links from edge node measurements. The problem of estimating link loss probabilities for a multicast distribution tree is examined. Our model assumes loss statistics are distributed to session participants by a network protocol such as RTCP. We propose a decentralized algorithm for ML estimation of the link loss probabilities in a chain of nodes rooted at the source node of the multicast distribution tree and terminating at a given leaf. An expression for the Cramer-Rao bound and an approximate form for the probability distribution function of the estimator are given. The performance of the algorithm is evaluated using computer simulations for a bottleneck detection application.