Light in fibre travels at about two-thirds of c, so every 100 km of path costs roughly 1 ms of round-trip time. Here is the physical floor under your ping.
You can buy more bandwidth, a better router and a faster plan, and the ping to a server on another continent will barely move. That is not a bug in your connection. It is physics: a signal needs time to cross the distance, and no upgrade makes it cross faster than light can. Here is the arithmetic behind that floor, why real ping always sits above it, and what it means for speed tests, VPNs and satellites.
Key Takeaways
- Light in glass fibre is slower than light in a vacuum. APNIC's networking blog gives approximately 194,865 km/s for fibre, against 299,792 km/s in a vacuum, roughly two-thirds of c.
- Rule of thumb (derived arithmetic, not a measurement): about 1 ms of round-trip time for every ~100 km of one-way fibre path. That is a minimum.
- Real ping is always higher than the floor. Cables do not follow straight lines, routes detour, queues add delay, and the last mile has its own cost.
- Only distance sets the floor. Bandwidth, router quality and plan speed change throughput, not how long light takes to travel.
- You can compare your own ping against the floor with the network speed test: a result far above it points at the path or at queueing, not at the speed of light.
The floor: how fast light travels in fibre
In a vacuum, light covers 299,792 km every second. In glass it is slower. APNIC's article on satellite and fibre delay states that the speed of light in glass or fibre-optic cable is approximately 194,865 km/s, which is about 65% of the vacuum figure, or "some two-thirds" of c.
That number turns directly into a delay per kilometre. At about 194,865 km/s, light covers roughly 195 km every millisecond, so 1 km takes about 5 microseconds one way. Everything below is derived from this single figure.
The rule of thumb: about 1 ms per 100 km
Ping is a round trip: the packet goes out and the reply comes back. So the distance is travelled twice.
- One-way delay = path length ÷ 194.9 km per ms.
- Round-trip time (RTT) = 2 × that.
- So 100 km of path costs about 100 ÷ 194.9 × 2 ≈ 1.03 ms of RTT.
Applied to a few well-known great-circle distances (approximate figures, straight-line, which no real cable follows exactly):
| Route | Straight-line distance | Minimum one-way | Minimum RTT |
|---|---|---|---|
| Same metro area | ~50 km | ~0.3 ms | ~0.5 ms |
| London to New York | ~5,570 km | ~29 ms | ~57 ms |
| London to Sydney | ~17,000 km | ~87 ms | ~175 ms |
These are floors computed from the formula above. They are labelled as derived arithmetic on purpose: they are what physics allows, not what any particular ISP delivers. If your London-to-New-York ping is 40 ms, something is wrong with the numbers, because the floor says it cannot be below about 57 ms on a fibre path of that length.
For the precise meaning of the quantity being measured, see the IETF definition of round-trip delay in RFC 2681, and of one-way delay in RFC 7679. MDN's latency glossary entry gives the browser-side view.
Why real ping is higher than the floor
The floor assumes one straight fibre between the two points and nothing else. Real paths add delay in four ways:
- Cables are not great circles. Fibre follows coastlines, rail corridors and seabed routes, so the path is longer than the straight line.
- Routes detour. Traffic follows business and peering agreements, not geometry; a packet between two nearby cities can travel via a distant exchange point.
- Queues add delay. Each router can hold packets briefly, and an oversized buffer holds them for a long time. See bufferbloat for how a busy link inflates ping far above the idle value.
- The last mile has its own cost. Wi-Fi, cable, DSL and mobile radio each add processing and scheduling time before the packet reaches the fibre.
Per-country or per-ISP "typical" numbers would just be guesses, so the useful habit is simpler: work out the floor for your distance, then treat the gap between the floor and your measured ping as the part that can be improved.
Satellites: a different kind of distance
A geostationary satellite sits roughly 36,000 km above the equator (about 35,786 km), and the signal has to go up and come back down. APNIC's article calculates that for a ground station at 45° latitude on the same meridian, the time for a signal to pass from Earth to the satellite and back is approximately 260 ms. That is a single up-and-down hop, before a single router on the ground has touched the packet. A request and its reply each make that hop, so a ping over a geostationary link starts at roughly twice that, over half a second. Low-earth-orbit constellations fly much closer to the ground, so the physical floor is far lower, but the same rule applies: the path length divided by the speed of the signal is the minimum.
What this means in practice
- Pick the nearest speed-test server. A distant test server measures the distance as much as your line, which is why the same connection can show very different latency from run to run. See how speed tests choose and measure and why results vary.
- A far VPN exit adds a floor of its own. The detour through the VPN server lengthens the path in both directions, and no protocol tuning removes that. The full breakdown is in why a VPN slows your connection.
- Latency is independent of bandwidth. A faster plan does not shorten the trip. The relationship between the two is covered in internet speed metrics explained, and the variation in latency is jitter.
- RTT gives an upper bound on distance, not a location. Because nothing beats the floor, a 10 ms ping means the other end cannot be more than roughly 1,000 km of fibre away. Queueing and detours make the real distance shorter than that bound, so latency can rule places out but cannot pin one down. Most IP location data comes from registries and databases rather than timing, and it has its own error, covered in how accurate IP geolocation is.
Compare your own ping with the floor
BrowserInsight's network speed test measures latency as the time your browser takes to fetch a tiny response from the test endpoint, averaged over several requests. The endpoint runs on Cloudflare's network, which normally answers from a data centre near you, so a low single-digit or low double-digit result is mostly your last mile and local network. That is useful in itself: if this nearby latency is already high, the problem is Wi-Fi, the access line or a busy link, not distance.
The same page also times requests to several large websites. Those numbers include more than the light-speed floor, because each is a full HTTP request from the browser and the site's nearest server may be far away, but the comparison still helps. If one site is far slower than the others, the path to it is long or detoured. Run the test a few times and compare with the floor for the distance you expect; the tool reports latency, not server locations, so the distance estimate is yours.
FAQ
Can a better ISP or fibre plan lower my ping to a distant server? Only down to the floor. A better route, less queueing or a wired connection can close the gap, but nothing can beat the distance divided by the speed of light in fibre.
Is 1 ms per 100 km exact? No. It is derived from approximately 194,865 km/s and rounds to about 1.03 ms. It ignores equipment delay and assumes the path is as short as the straight line.
Why is satellite ping so much higher? On a geostationary link the signal travels roughly 36,000 km up and the same back down. In the APNIC example that one hop alone takes around 260 ms, and a ping needs a hop each way, so it starts above half a second before any ground delay is added.
