Servers in 16 cities ping each other, 16 packets, twice. The fastest reply counts.
Each pair is routed over the in-service submarine cables and straight land links, separately on each map.
A reply faster than light in glass along a map's route rules that map out for that pair.
Each pair is routed over in-service submarine cables and straight overland links, separately on each model.
A round-trip time below the fibre propagation floor along a model's route is inconsistent with that model.
Results by map
+Results by model
Speed limits
+Propagation limits
Light in a vacuum travels 299,792 km/s. In standard single-mode fibre (group index 1.468) it travels 204,190 km/s. Every submarine cable is fibre, so the minimum round trip is 2 × route length ÷ 204,190 km/s. That floor sets each bar and the speed of the gold pulses on the map.
-Measured times sit above the floor: installed cable is longer than its drawn route, and routers and queues add delay. A time above the floor proves nothing. A time below it means the map puts the cities too far apart.
-Light in a vacuum is 47% faster than light in fibre; radio through air is within 0.03% of it. A time below the vacuum floor is impossible for any cable, radio link or satellite.
+Light travels at 299,792 km/s in vacuum and at 204,190 km/s in standard single-mode fibre (group index 1.468). Submarine cables are fibre, so the minimum round-trip time over a route is 2 × route length ÷ 204,190 km/s. This floor determines each bar and the speed of the animated markers on the map.
+Measured times exceed the floor because installed cable is longer than its charted route and because routers and queues add delay. A time above the floor is therefore uninformative. A time below it implies that the model places the endpoints too far apart.
+The vacuum speed exceeds the fibre speed by 47%, and radio propagation through air is within 0.03% of the vacuum speed. A time below the vacuum floor cannot be achieved by cable, radio or satellite.
Speed each live route would need
Implied propagation speed
How each map measures distance
+Distance on each model
Every measurement
+All measurements
-Log scales, milliseconds. Small dots: WonderNetwork's public 2020 ping set. Large dots: the live grid.
+Cable crossings
-RIPE Atlas anchor-to-anchor traceroutes. The jump in round-trip time at the sea-crossing hop is the cable itself, compared with light along that cable's drawn route.
+RIPE Atlas anchor-to-anchor traceroutes. The increase in round-trip time at the hop that spans the sea isolates the cable segment; it is compared with the fibre floor along that cable's charted route.
-Bars: light in glass along the cable's route, there and back. White line: the measured jump, fastest of a day of traceroutes every 15 minutes. Click a crossing to see the route on the map.
+Bars: round-trip fibre floor along the cable's charted route. White mark: measured increase, minimum over 24 hours of traceroutes at 15-minute intervals. Select a crossing to show its route on the map.
Check your own ping
-Ping a server from your computer and enter the lowest time. Local network and ISP delay only add time.
+Networks on the measured paths
+Every endpoint and traceroute hop was attributed to the autonomous system (AS) that announced its address in the public BGP routing table on 23 October 2026 (RouteViews route-views2, via CAIDA prefix2as), named from the regional internet registries' AS list (RIPE NCC). Addresses absent from the table were matched to internet exchange point prefixes registered in PeeringDB. Each hop is a router of an identified network, and adjacent networks exchange traffic only under peering or transit agreements configured by both parties. The attribution can be repeated from the cited public sources.
+ +Measurement endpoints
+RIPE Atlas anchors
Carriers on the Vultr traceroutes
Internet exchange points
Sea crossings: networks either side of the jump
+Best attainable flat arrangements
+Latency bounds distances only from above, so a free flat arrangement can satisfy any set of measurements by shrinking. The informative question is how far a given flat map would have to be corrected. Each city may move at most the stated distance from where the map places it; the arrangement shown minimises the total excess over all measured pairs at the speed of light in fibre, using straight lines (the shortest possible cable). The final setting is the smallest correction at which no route remains inconsistent. The problem is convex, so this minimum is global.
+ + +Independent verification
+To reproduce a measurement, ping one of the anchors from any computer and enter the minimum round-trip time. Local network and access delays can only increase the measured value.
Code and data
+The code, the raw measurement logs and the derived data are public. Code is released under the MIT licence; the measurements and derived data under CC BY 4.0. Third-party inputs remain under their own terms and are downloaded by a script rather than redistributed.
+ + +