New Metric stellt die Hälfte der Geolokalisierungsdatenbanken als übertrieben aus

Over half of paths from DB-IP and IP2Location geolocation databases need substantial correction when checked against actual network physics. That's the finding from a new paper introducing Path Consistency Scoring (PCS), a passive framework that treats router geolocation as a path-level consistency problem instead of trusting the point coordinates databases hand you.

PCS Treats Geolocation as a Sequence Problem, Not a Lookup

PCS models each traceroute as a sequence of candidate city-level locations. A Hidden Markov Model fuses local metadata - rDNS labels, Geofeeds, IXP data - with speed-of-light constraints and empirical RTT latency priors. The output is a single path consistency score: how well the metadata and observed RTT increments support a coherent geographic interpretation.

That score only matters when latency actually proxies for geography. So the authors also define Path-Model Alignment, which compares speed-of-light residual increments of the decoded path against a reference path from active probing. This alignment metric is where the ugly truth about database quality comes out.

413,354 Traceroutes: Databases Are Not All Equal

The team evaluated on 413,354 RIPE Atlas traceroutes, with a 6,555-path subset verified by active probing. On those validated paths, 94.2% of decoded sequences achieved mean error below 200 km. PCS is largely GeoDB-agnostic - median scores vary by less than 5% across four commercial databases - which means the consistency signal is robust, not database-dependent.

But the alignment metric tells a different story. Over half of DB-IP and IP2Location paths require substantial correction, compared with just 15% for IPinfo. That's not a subtle difference. If you're running a measurement study and you don't check alignment, you're inheriting database accuracy - or inaccuracy - without qualification.

What This Means for Downstream Network Analysis

PCS gives researchers a calibrated confidence score for every geographic path. Instead of blindly accepting a point coordinate, you can now ask: does this path even make physical sense given speed-of-light round-trip times? The answer for a lot of commercial databases is "barely."

Expect this framework to become a standard filter in large traceroute analyses. When your conclusions about intercontinental latency or peering geography depend on router location, PCS tells you which hops to trust and which to discard, without needing active probing on every path.

Source: Overconfident Coordinates: Quantifying Confidence in Traceroute Geolocation
Domain: arxiv.org