Statistical Geolocation of Internet Hosts

Automated geolocation of IP addresses has important applications to targeted delivery of local news, advertising and other content over the Internet. Previous measurement-based approaches to geolocation employ active probing to measure delays among a set of landmark nodes with known locations. The location of a target IP address can be approximated by that of the nearest landmark, as determined by the delay measurements. To improve geolocation accuracy, a variation of this approach uses multilateration with geographic distance constraints to obtain a continuous location space rather than the discrete set of landmark locations. Since the previous approaches are fundamentally deterministic, they can only provide relatively loose bounds on the true location of an IP address. We develop a statistical geolocation scheme based on applying kernel density estimation to delay measurements among a set of landmarks. An estimate of the target location is then obtained by maximizing the likelihood of the distances from the target to the landmarks, given the measured delays. This is achieved by an algorithm which combines gradient ascent and force-directed methods. We present experimental results on PlanetLab to demonstrate the superior accuracy of the proposed geolocation scheme compared to previous methods.