GeoLink: a Multimodal approach Internet Scale Geolocation System for Improved Coverage and Accuracy

Despite being critical infrastructure, it remains challenging to accurately position the physical location of Internet routers. In contrast to end-host geolocation where provider databases, registry information, and crowd-sourcing techniques can be successfully leveraged, router geolocation is known to be relatively inaccurate. While concerted efforts by the research community have produced recent systems capable of significantly better accuracy, their coverage is generally low; a state-of-the-art systems is capable of geolocating only 7.6% of the router IP addresses in Internet scale topology datasets. We develop, evaluate, and make public the GeoLink system for geolocating router IP addresses at Internet scale. Our key insights are three-fold: i) utilize unique capabilities and properties of routers, i.e., timestamps, aliases, and DNS names; ii) use multiple active measurement techniques to derive per-link latencies and identify co-location; and iii) combine existing techniques to improve coverage and refine candidate inferences. To bootstrap GeoLink, we extract geolocation candidates from multiple data sources including existing geolocation databases, DNS names, and PeeringDB. We finally use our new measurement techniques to obtain sound link latencies, extend coverage for bidirectional links, and ensure aliases are location geo-consistent. On a large validation set, GeoLink provides 5 times coverage improvement over existing systems, while simultaneously lowering error.

 

Short bio: 

Kevin Vermeulen is a researcher at CNRS. He has obtained his doctorate at Sorbonne Université ok internet cartography in 2020 and did a postdoc at Columbia University on building systems to analyze and improve internet paths before joining the CNRS. He is interested in building measurement systems for a better understanding of the internet.