Least squares estimates of network link loss probabilities using end-to-end multicast measurements

| Estimation of quality of service parameters associated with large-scale computer and communication networks is a problem of considerable importance. In this paper, we consider estimation of link-level loss probabilities based on active tomography using multicast probing schemes. We formulate a regression framework for the problem, develop and study the properties of several types of least-squares based estimators. These include ordinary, generalized, and iteratively reweighted least squares estimators. We study the asymptotic and nite-sample properties of these estimators. The rst two are simple to compute while the last two are asymptotically eÆcient. Computation of the variance-covariance matrix and inference using these estimators are much simpler computationally than those based on the MLE and E-M algorithm.