Adaptive Constraint Based Geolocation

Abstract

Location-based web applications rely on e ective Geolocation of Internet hosts. Measurement-based geolocation techniques utilize reference hosts, termed landmarks, having a known geographic location. Previous techniques either use a single constraint per landmark or use trace-routes to form addition topology constraints causing massive network tra c. In contrast, we propose Adaptive-Constraint-Based Geolocation (ACBG), which uses multilateration with two distance constraints per landmark to deduce the geographic location of Internet hosts. ACBG requires just a single delay measurement per landmark hence minimizing network tra c. However, to form these distance constraints an accurate method of delay to distance translation is required. This is a challenging problem since the translation between network delay and actual geographic distance is e ected by many factors (such as indirect path and queuing delays) and also varies from region to region. ACBG accurately uses network delays to form distance constraints, and then uses multilateration to geolocate the target host. Practical testing results using three types of landmarks i.e. PingER, Planet Lab and PerfSONAR, show that ACBG outperforms previous measurement based techniques with overall mean error reduction of 26% and 90th percentile error reduction of 37%.