Robust Statistical Geolocation of Internet Hosts

Measurement-based approaches to geolocation of Internet hosts typically employ active probes sent from a predetermined set of landmarks to the target with unknown location. A significant improvement of such approaches can be made by constructing kernel density estimators using measurement data of end-to-end delays among all landmarks within the set. However, obtaining the data for this statistical approach to geolocation can be time-consuming, whereas the measurements are not all of high quality due to the time-varying characteristics of the network. To overcome these drawbacks, an ordinal optimization scheme is proposed to determine a subset of the landmarks that will yield improved geolocation accuracy with substantially less overhead. Given a set of kernel density estimates obtained by the landmarks, an optimization-based approach, with differential evolution as the search engine, is developed to find the location of the target. Numerical results demonstrate the efficiency, accuracy, and robustness of the proposed geolocation scheme.