Large-scale, discrete IP geolocation via multi-factor evidence fusion using factor graphs

Traditionally, IP geolocation has been focused on finding the exact location of a given internet host. In this work, we focus on the problem of simultaneously geolocating large numbers of IP hosts to within a discrete set of geographical regions. Towards this objective, we present a fusion algorithm that combines information from multiple, heterogeneous sources of location information. Our algorithm estimates the most likely region for all hosts simultaneously. A key component of the algorithm is a systematic procedure for automatically constructing factor graphs to perform information fusion from a collection of delay measurements between hosts. While the paper focuses on the algorithmic aspects of the problem, we present initial results that demonstrate that hosts are placed in the correct region with very good accuracy.