Geolocation of devices has long been a topic of research, and using the network to help uncover these details is a specific area within the broader topic. There have been several efforts to use information on and about the Internet to help systematically determine the physical location of a given machine. The authors present new work in the area of Internet protocol geolocation based on end-to-end delay measurements, work that performs better than previous efforts, including those based on domain name system, traceroute, and ping.

The authors present a constraint-based geolocation (CBG) approach that uses the additive delay error in end-to-end delay measurements to estimate a great-circle radius. Multiple great-circle radii from a number of landmarks are then used to perform multilateration to create an area of estimation within the intersection of all radii. The centroid of this enclosed polygon is then used as the estimated location, with the area of the polygon representing an estimate of accuracy. The bounding region is reduced by a distributed self-calibrating process, whereby each of the landmarks continuously adjusts its estimation delays and tunes the distance calculations based on changing conditions within the network.

Based on tests in Western Europe, the US, and via PlanetLab, the authors show improved results and analyze the impact of various changes such as the number (and quality) of the landmarks chosen for the geolocation task. The median error for the tests in the US is below 100km, while those performed in Europe have a median below 25km. This is an improvement over the median of 150km for the US and 100km for Europe using methods such as GeoPing.