Every host selling unmetered traffic has a number in mind. Most of them will not tell you what it is, which means the number is whatever their upstream bill tolerates on the month they decide to look at your port. Ours is below, and it has been the same since we published it.
The figures
Fair use is expressed as a sustained average over a rolling thirty days, not a monthly cap. Bursting to line rate is expected and is the reason you bought the port.
| Port | Sustained fair-use average | Roughly, per month | Burst |
|---|---|---|---|
| 10 Gbit/s | 1.5 Gbit/s | About 480 TB | To line rate, unrestricted |
| 25 Gbit/s | 4 Gbit/s | About 1.3 PB | To line rate, unrestricted |
| 40 Gbit/s | 8 Gbit/s | About 2.6 PB | To line rate, unrestricted |
Inbound and outbound are counted together. Traffic between two of your own instances at the same site is not counted at all, because it does not touch transit and charging you for it would be an invention.
What happens if you cross it
You get an email. Not an invoice, not a throttle, not a suspension notice written in the passive voice.
The email arrives when the thirty-day average passes the figure above, states the average we measured and the window it covers, and asks what you are doing. Roughly a third of the time the answer is that something is misconfigured and the customer fixes it that afternoon. Another third are genuinely growing, and the conversation becomes a dedicated 40 Gbit/s port at forty-five euro a month, which is cheaper than what they were about to be charged elsewhere. The remainder are running a workload that belongs on a storage node with a fat port rather than a general instance, and we say so.
If none of that goes anywhere, we shape the port to the fair-use rate and tell you the day we do it. Nobody has ever received a surprise bandwidth invoice from us and nobody is going to, because we do not have a mechanism for producing one.
Attack traffic never counts
Volumetric traffic arriving at your address during an attack is filtered at the edge, up to 12 Tbit/s at the largest sites, and none of it is counted against you. This ought to be obvious. It is not universal, and the reason we mention it is that two customers have arrived here after being billed by a previous provider for traffic that was aimed at them by somebody else.
Details of how the filtering works are on the network page; the short version is that it is always in path, so there is no detection delay and no button anyone has to press.
Why the industry keeps it vague
Three reasons, in descending order of honesty.
The first is genuine: transit is committed in advance, and a provider with thin margins cannot promise a number they might not be able to buy. That is a real constraint and it deserves sympathy. It is also solvable by publishing a smaller number.
The second is that vague terms are a commercial instrument. If nobody knows the limit, nobody self-selects out, and the heavy users can be identified and repriced individually later, at leisure, from a position of strength.
The third is that a lot of hosts do not know their own figure. They have an aggregate transit bill, a rough sense of which customers are expensive, and no per-port accounting worth the name.
The 40 Gbit/s port
Dedicated forty-gigabit ports became an option this year on the EPYC, storage and bare metal lines, at sites with the upstream capacity to back them. Forty-five euro a month, unshared, no contention with the rack.
They exist mostly for two workloads: storage nodes that need to drain a hundred terabytes at a sensible rate, and anyone doing egress-heavy distribution where the sustained average on a ten-gigabit port was always going to be the constraint. If you are considering one because you like the number, the ten-gigabit port is almost certainly correct and we will tell you so before taking the money.
The one thing we will not do
We do not sell unmetered and then define fair use retroactively in a support ticket. The figures above are in the terms, they change with thirty days notice and never retroactively, and if we ever have to lower them the announcement will be a post in this journal with the transit economics attached.