From Timed Pass Rate to Estimate Delay Distribution

Delay tomography and loss tomography are two key components of network tomography, the former is for estimating link level delay distribution of a network, while the latter for estimating link-level loss rates of a network, by end-to-end measurement. The difference between the objectives leads to two independent studies and creates two types of methods for one each. Preti et. al. attempted to connect these two together and borrow some techniques used in loss tomography to delay tomography. The attempt achieved its success in terms of performances but also unveiled its weakness in terms of statistical efficiency. In this paper, we extend the previous work to overcome the weakness,where an expectation and maximization procedure is added that considers all information obtained by receivers. The results show the new method performs significantly better than the previous method without scarifying its performance.