Summary

  • At RIPE 93, Katherine Izhikevich described an Atlas probe still labelled Germany after its host moved it to Brunei. It had reported roughly 20 ms to German servers, then roughly 30 ms to Singapore; her team's minimum physical bound for Germany to Singapore was about 100 ms.
  • The team found a conservative minimum of 2% of Atlas vantage points with impossible reported locations during January 2019 to April 2026. Its outreach prompted 93 corrections, while ten hosts denied a reported problem.

The pin stayed; the probe left

The probe went offline and came back with an impossible itinerary. Before the break, its round trips to German servers sat near 20 ms. Afterward, it reached Singapore in about 30 ms while its public location still said Germany. Izhikevich told the RIPE 93 MAT session that the host had moved the probe to Brunei without changing the location record. Brunei is near enough to Singapore for the new timing. Her team's physical lower bound for Germany to Singapore was around 100 ms. The timestamped packet contradicted the static map label.

That is a problem for anyone using RIPE Atlas as a geographical witness. A researcher may select a probe for its declared country, then interpret latency or routing through the wrong continent. Izhikevich's official talk abstract reports that at least 2% of vantage points had impossible operator-reported locations at some point from January 2019 to April 2026. The test catches the glaring cases. It cannot count a probe moved a few streets away, and the team's public result is a lower bound for the period, not a present-day census.

Check the ruler before blaming the pin

Her method used Atlas's regular measurements toward DNS root servers. The responding root instance gave a comparison location; the measured round-trip time had to clear a minimum possible time set by distance and the speed of light. A 30 ms Germany-to-Singapore round trip failed that test. In the same session Q&A, an operator asked whether the root instances themselves were correctly located. The team cross-checked and found one likely wrong instance. A faulty ruler could falsely accuse the probes measured against it, so the team examined the reference before treating its flags as evidence.

The repair has a human end. Izhikevich reported 93 hosts updating locations after outreach and ten denying the report. Software probes were about three times as likely as hardware probes to breach the speed bound, she said; moving a virtual machine can leave its owner chasing a stale coordinate. RIPE Atlas's geolocation guide tells hosts how to move a location and lets others dispute a badly placed probe. Its small public-map privacy offset is a different matter from a Germany-to-Brunei move.

The researcher's original repository publishes regularly updated lists of likely mislocated probe IDs. A study that selects locations from Atlas can check that list and record the probe's location history before drawing a country-level conclusion. The map is useful because many hosts keep it accurate. Izhikevich showed what happens when one does not: a result can be measured precisely and still be plotted on the wrong continent.

Sources: RIPE 93 talk; MAT transcript; RIPE Atlas location guide; original research repository.