CDN origin servers.
Pull as hard as you like.
The machine your edge fleet pulls from, on a port nobody is counting. Cache misses, purges and release days arrive all at once, and that is the traffic an allowance handles worst.
- port
- 10 to 100 Gbps
- egress
- unmetered
- locations
- AMS · MEP · IAD · PHX
What an origin is, and what it has to survive.
A CDN keeps copies of your files on edge servers close to your users. When an edge server does not have a file, it asks your origin: the one machine that holds the original. The origin is quiet for most of the day and busy at the worst moments, after a purge, after a release, when a new region goes live.
That shape of load decides what to buy. You are not sizing for the average hour. You are sizing for the minute when every edge node asks at once, which is why the port and the disks matter more here than clock speed.
How to choose.
- 01 ─ peak
Find your worst day, not your average. Take the biggest purge or release from your CDN logs: the origin has to answer that, not a quiet Tuesday.
- 02 ─ hot set
Work out which files the edge asks for most. If they fit in RAM, most pulls never reach a disk.
- 03 ─ port
Turn the peak into Gbps and leave room for the next edge node you add. The calculator further down does the arithmetic on your numbers.
- 04 ─ location
Put the origin where your edge fleet pulls from. The shorter the path, the shorter a cache miss.
- 01 ─
Release days arrive all at once
on a meter or a shared hostA purge or a release sends the whole edge fleet back to the origin together. That burst is what a transfer allowance prices worst.
hereThe port has no counter. The busiest day of the month costs the same as the quietest.
- 02 ─
A hit ratio that wanders
on a meter or a shared hostOn a shared host a neighbour competes for the same page cache, so how much of the working set stays in RAM moves under load.
hereSingle tenant: the RAM and the page cache are yours, and the hit ratio moves only with your own traffic.
- 03 ─
The fleet outgrows the port
on a meter or a shared hostMore edge nodes pulling at once turn a narrow uplink into the bottleneck before the disks or the cores are.
here10 Gbps on the stock plans, 25, 40 or 100 Gbps as a custom build.
What to look for in the machine.
An origin answers every edge node's misses at once. A narrow port is the first thing a release saturates.
A purge turns hits into misses, and misses land on disk. NVMe keeps that queue short.
Every object served from RAM is a read the disks never see. Size it to what the edge asks for most.
Thousands of concurrent pulls and TLS handshakes spread across cores; one fast core does not carry them.
The servers that fit this today are marked under the processor on the price list. The shelf changes; what to look for in the machine does not.
What reaches the origin.
Set what the CDN serves and how much of it the cache answers. The origin carries the rest.
- origin egress
- 5 Gbps
- each point of hit ratio
- 1 Gbps
- a month at that rate
- 1,620 TB
- billed for it
- $0
carries over: port 10 Gbps
Questions to settle before ordering.
- 01 ─
Can I run origins in more than one location?
Yes, in any of our four locations: AMS, MEP, IAD, PHX. Each location has its own set of builds in stock; the unmetered port is the same everywhere. Each server has its own port; capacity is per machine and cannot be pooled across servers (Terms, section 3).
- 02 ─
Do you shape or throttle what the origin serves?
No. We do not rate-limit below the port you ordered, and we do not shape traffic by protocol, destination or content (Terms, section 3).
The rest, for every machine: all questions and the terms.
- We are not a CDN. We sell the box your CDN pulls from: no edge nodes, no anycast, no cache tuning.[no]
- No object storage API: it is a server with disks, and what speaks HTTP on top is yours.[no]
- No mitigation product if the origin address leaks.[no]
Tell us what it serves.
The workload, the working set, where the users are. We come back with a machine sized to it, a price and a date.
alsounmetered servers · streaming servers · game servers · storage servers · download servers · database & compute · virtualisation hosts · blockchain node servers