Paris → Frankfurt · ~12 ms
A route nobody calls slow.
One hop off the European backbone
“One country” only costs you something if that country is far away. It is not — and this page is the arithmetic rather than the assurance.
- Bucharest
- 12 ms
- Frankfurt
- 35 ms
- Amsterdam
- 42 ms
- Method
- Median, 30 days
Measured
Twelve cities, sorted by distance
Sorted rather than grouped by region: you stop reading as soon as you have passed your own city.
- Bucharest 12ms
- Kyiv 21ms
- Warsaw 29ms
- Vienna 33ms
- Frankfurt 35ms
- Istanbul 38ms
- Amsterdam 42ms
- Milan 45ms
- London 48ms
- Stockholm 52ms
- Dubai 92ms
- New York 118ms
Highlighted cells are under 50 ms — the range inside which SSH, remote desktops and interactive work feel local. Every European city on this list is inside it except Stockholm, which misses by two.
For scale
Thirty-five milliseconds is not a compromise
A number only means something next to one you already know. Here are three routes nobody calls slow, and where ours sits among them.
Chișinău → Frankfurt · ~35 ms
Ours. Three times that, and still under the threshold where a human notices.
London → New York · ~75 ms
Crossed by every trading desk in the City without complaint.
Frankfurt → Singapore · ~160 ms
Where interactive work genuinely starts to hurt.
How it was measured
A latency figure without a method is an adjective
Anyone can publish a good number. What makes it worth reading is knowing how it was produced — and being able to produce it yourself.
ICMP from our edge, not from a probe network.
Measured from the same routers that carry your traffic. Third-party probe networks sit in datacentres with better peering than the customers they claim to represent.
Median over 30 days, not a best run.
One packet at three in the morning proves nothing. The figure is the median of continuous sampling, so a good night cannot flatter it.
To the city, not to the nearest peering point.
Targets are well-known hosts inside each metro, not the edge router of whichever exchange happens to be closest to us.
And you can repeat it.
Every target is public and the looking glass runs the same commands from our side. If your number disagrees with ours, we would like to know.
Before you order
Three questions the numbers raise
If your own measurement disagrees with this page, tell support — we would rather correct a figure than defend one.
Open the looking glassMy users are in the Americas. Is this viable?
For a website with a CDN in front, entirely — the origin latency stops mattering once assets are cached at an edge. For an interactive service with players or traders in New York, 118 ms is real and you should feel it before you commit. Ask support for a trial and measure it.
Does the DDoS filtering add latency?
Under a millisecond in steady state — the filtering is upstream and on-path, not a diversion through a scrubbing centre elsewhere. Under active attack you may see a few more, which is a trade every alternative also makes, usually without saying so.
Why is Istanbul slower than Frankfurt?
Because geography is not topology. Traffic to Istanbul currently transits through Bucharest and Sofia rather than crossing the Black Sea, which adds distance a map does not show. A third carrier would likely improve it, and we will republish the figure when it does rather than quietly leave the old one up.