Octant: a comprehensive framework for the geolocalization of internet hosts
Bernard Wong, Ivan Stoyanov, Emin Gün Sirer · eCommons (Cornell University) · 2007
Determining the physical location of Internet hosts is a critical enabler for many new location-aware services. In this paper, we present Octant, a novel, comprehen-sive framework for determining the location of Internet hosts in the real world based solely on network mea-surements. The key insight behind this framework is to pose the geolocalization problem formally as one of error-minimizing constraint satisfaction, to create a sys-tem of constraints by deriving them aggressively from network measurements, and to solve the system geomet-rically to yield the estimated region in which the target resides. This approach gains its accuracy and precision by taking advantage of both positive and negative con-straints, that is, constraints on where the node can and cannot be, respectively. The constraints are represented using regions bounded by Bézier curves, allowing pre-cise constraint representation and low-cost geometric op-erations. The framework can reason in the presence of uncertainty, enabling it to gracefully cope with aggres-sively derived constraints that may contain errors. An evaluation of Octant using PlanetLab nodes and public traceroute servers shows that Octant can localize the me-dian node to within 22 mi., a factor of three better than other evaluated approaches. 1