Unmetered servers.
Any port, 1 to 100 Gbps.
Nothing counts the traffic on any port we sell. You choose the port: up to 10 Gbps it is on the shelf, above that we build the machine around it.
- traffic
- unmetered
- ports
- 1 · 10 · 25 · 40 · 100 Gbps
- on the shelf
- up to 10 Gbps
- built to order
- 25 · 40 · 100 Gbps
- locations
- AMS · MEP · IAD · PHX
- 01 ─
The bill follows the traffic
on a meterAllowances, per-gigabyte tiers and 95th percentile turn a busy month into a bigger invoice.
hereNo counter on any port. The invoice is the same number every month.
- 02 ─
Ordinary operations get billed
on a meterRetries, monitoring scrapes, backup reads and cache re-warms count against an allowance like the traffic you meant to serve.
hereRun the monitoring and the backups you need; none of it becomes a line item.
- 03 ─
A limit hiding in the small print
on a meter“Unlimited” plans often carry a fair-use clause that brings the cap back on the day it matters.
hereNo allowance, no fair-use clause, nothing in the terms that brings a cap back. The port is yours alone. The uplink behind it is shared, sized for the servers on it, and when it gets close to full we add capacity rather than hold anyone back.
How allowances and the 95th percentile are counted, with a graph you can play with: how traffic is billed.
Same terms on every step. The machine changes.
Flat monthly and nothing counted, from 1 Gbps to 100. Going up, the question stops being the port and becomes the machine that can fill it.
A big port needs a machine that can fill it.
Fitting a faster adapter is the easy part. 25, 40, 100 Gbps are real steps, not one jump: if 25 or 40 is what the workload fills, that is what we build.
NIC interrupts and the kernel's network stack need cores before the port saturates. A four-core box does not get a 10 Gbps adapter.
Window scaling, round-trip time and the far end cap a single TCP stream well below line rate. Edge pulls, replication and concurrent downloads fill a big port; one stream rarely does.
100 Gbps is about 12.5 GB/s, more than a single NVMe drive sustains. The array is sized to keep the port fed.
The drives and the NIC compete for the CPU's lanes, and one quietly starves the other. The CPU is chosen for its lane budget, not only its core count.
A GPU is available on a built-to-order machine.
Which port your peak needs.
Set what you send at the busiest moment. The rows show which port carries it and whether that is a stock server or a build.
- smallest port
- 10 Gbps
- where it comes from
- stock server
- that peak, all month
- 972 TB
- billed for it
- $0
- Lawful traffic only. The acceptable use section is short, specific and enforced.[limit]
- No DDoS mitigation product at any port size: filtering at the network edge and monitoring, nothing sold beyond that.[no]
- No 10 Gbps port on a small build: a machine that cannot fill it would be paying for a number on an invoice.[no]
- No click-to-order path to 100 Gbps. Builds above 10 Gbps are assembled to order and take longer than the stock range of a few hours to 7 business days.[limit]
Up to 10 Gbps, pick a server. Above it, send the spec.
1 and 10 Gbps are on the price list with $0 setup. 25, 40, 100 Gbps come back with a price and a date.