Methodology

Where the measurements come from

RIPE Atlas anchors (servers RIPE places in data centres) are traceroute targets for every other anchor in the world and for a few hundred probes near each one, every 15 minutes, and the results are public. We read the newest result once an hour from every anchor and probe in the metros we follow, so about one round in four is kept. Probes registered without a real location (they sit on the "centre of the US" default point in Kansas) are left out.

How a path is judged

Time beyond the source's own line

The destination's round trip less that of the first public hop before it. A home line can take several milliseconds before anything else; a data centre's first hop takes a fraction of one. When no public hop answers before the destination, nothing is taken off.

Left the metro

7 ms or more beyond the source's line. Light in fibre takes about 1 ms per 100 km there and back, so no path inside one metro takes that long (Kansas City to Chicago is 660 km: at least 6.6 ms). The hop from which the times stay above 7 ms is shown; one slow hop on the way isn't evidence, since routers answer traceroute slowly when busy.

Local, via an exchange

Under 7 ms, with a hop inside an exchange's peering LAN (PeeringDB). The exchange is placed by that hop's own time, not its home city: STLIX's LAN reaches into Kansas City, so a STLIX hop at under a millisecond from Kansas City is Kansas City. An exchange based in another city is named with the building where PeeringDB lists it in the metro (STLIX and MICE: 1530 Swift); if it lists none there, the page says so, since the hop then shows only a member router's port on that exchange.

Local, no exchange hop

Under 7 ms with no exchange LAN hop: a private link or a shared upstream in the metro.

Unclear

The destination didn't answer, or the source's own line takes 7 ms or more (then where its first public hop is can't be told). Shown, never guessed.

One direction at a time

A traceroute shows the way out; its time includes the way back. A path can cross a local exchange going out and still take long because the reply came back another way. Each pair is shown in both directions where both are measured. RIPE's public measurements go from probes towards anchors; for the way back we run our own traceroutes from the anchors to the probes once a day (ICMP, at an hour that changes daily), within a credit budget: pairs whose way out left the metro first, then the ones measured longest ago. A home router that doesn't answer traceroute leaves its way back unclear.

Where an endpoint is

Registered locations can be wrong. An endpoint that answers anchors elsewhere faster than light in fibre could from where it's registered (for at least two of them, and more than any of them has such partners) is counted in the metro it answers from, and its page says why.

What the data can't show yet

Most sources today are hosting and data-centre networks, where the anchors are; few are home or business access networks. Network names come from PeeringDB, or the anchor's company when PeeringDB has none.