Multicast-Based Inference of Network Internal Loss from End-to-End Data

Loss tomography aims to obtain the loss rate of each link in a network by end-to-end measurements. If knowing the loss model of a link, we deal with a parametric estimate problem with incomplete data. Maximum Likelihood Estimate (MLE) is often used in this situation to identify the unknown parameters in the loss model, but it relies on the iterative approximation to identify the parameters that requires a long execution time. Pseudo Maximum Likelihood Estimation (PMLE), such as the bottom-up approach [7], is very fast for estimating link loss rates, but it does a poor job to estimate non-leaf links. In this paper, we propose an approach for estimating the loss rates of the links based on PMLE and MLE. In contrast to the previous methods, our approach can improve the efficiency greatly without any reduction of the accuracy. Simulation results show that the inferred results accord with the theoretical loss rates of the links.