Method

What is measured

For every location a provider publishes, crackthe.net records what the provider’s own server list says about it. For a rotating sample of locations, one published server endpoint is pinged from probes in many countries. Every reply is stored with the probe, time, and method version.

What latency can show

A reply that takes r milliseconds for the round trip cannot come from more than r × 100 km away. Light travels about 200 km per millisecond through fiber, and a round trip covers the distance twice. Actual routes are longer than a straight line, so the endpoint is usually much closer than this upper bound.

The check works in only one direction. A fast reply to a distant probe can rule a location out. A slow reply rules nothing out, and no reply proves a location.

Besides ruling a place out, a reply can also narrow down how far away the server can be. If a server is within r km of a probe and that probe is g km from the listed country, the server can be at most r + g km from that country. “Matches the country” means this distance is small. It never means “confirmed to be there.”

Rules

1. A probe rules a location out when it is farther from the stated country than its reply time allows. An additional 100 km is allowed for imprecise probe locations. If the provider publishes city coordinates, the same test is run against a circle with a radius of 50 km around them. The result for the city is shown next to the status and does not change it: a server in another city of the same country matches the country.
2. A location is shown as “server outside the country” only when at least 2 probes in different networks or cities rule the country out. If the provider does not mark the location as virtual itself, a second day, at least 12 hours later, must also show the same. Until then the status is “no finding.”
3. Probes whose stated location conflicts with the others are discarded. Two probes that both reach the same endpoint cannot be farther apart than the sum of their two distance limits. If they are, the probe involved in the most conflicts is discarded. A probe discarded this way in several measurement runs is no longer used in any run.
4. A location is shown as “matches the country” only when no probe rules it out and the closest reply places the server within 1,000 km of the listed country. If no probe is close enough, the result is “no conclusion possible.”
5. A single measurement run counts as an observation. The result is considered confirmed once a second day, at least 12 hours later, shows the same.

What is measured first

Locations are measured in turn, but not in arbitrary order. First come locations where a finding could say something the provider’s list does not: lists without a virtual marker, then unmarked locations in lists that do use such markers, and finally locations the provider marks as virtual. If a location’s addresses are close to those of a location the same provider lists in another country, probes in that country are queried as well. When a location is ruled out, the path to it is traced from the nearest probe. The last routers on that path that provide a name are shown with the result.

Networks

For every published endpoint, the site also records which network announces its address on the internet. This information comes from a public routing table, not from our own measurement. The country shown for a network is where the network is registered and says nothing about where a server is located. The table shows which providers place servers with the same operators.

Limitations

• The published server endpoint is measured, not the address the traffic exits through. Confirming the exit requires an account with the provider.
• The status is about the country. The city is checked only where the provider publishes coordinates; that result is then shown with the individual location. Regions within a country, such as U.S. states, are not distinguished.
• Most probes are located in data centers, and their locations come from IP geolocation.
• Country outlines use a scale of 1:50 million. Small islands near a coast may be missing.

Using the data

The figures on this site and the JSON files containing the same data may be quoted and reused with visible credit to crackthe.net and a link to the page they came from. The current state and changes from the last three months are published. The underlying measurement history is not.

Sources

• Server lists: each provider’s own public list, linked on the provider’s page.
• Probes: Globalping.
• Country outlines: Natural Earth, public domain. Country names: Unicode CLDR.
• Networks: iptoasn.com, public domain.