The geographical location of Internet IP addresses is important for academic research, commercial and homeland security applications. While commercial databases claim to have a very high level of accuracy, the correctness of their databases is questionable.
Academic tools, based on delay measurements, were shown to have a large range of error. We present a novel algorithm that crawls the Internet PoP level graph to improve the accuracy of geolocation, combining information from both
geolocation databases and delay measurements. We show that the results provided by the algorithm are more accurate than geolocation databases information while avoiding the pitfalls of delay measurements’ usage
(pdf)(slides with voice over)