Move a machine to another type at any time with Resize on the machines page. A bigger type is always possible; a smaller one only on cloud native compute.
Three machine types
When you configure a project you choose a machine type beside the datacentre. It decides how the disk works. Not every datacentre sells all three; where only one is offered, the question is not asked.
| Regular compute | Cloud native compute | |
|---|---|---|
| The disk | comes with the machine | a volume of its own |
| Where it lives | on the machine, local storage | on the provider's network |
| Choosing the size | whatever the type includes | yours, from 10 GB up |
| Growing it | move to a larger type — see It is downtime below | grow the volume, machine unchanged |
| Best for | most sites | files that grow unpredictably |
Bare metal is the third type: a whole physical server that is yours alone, at the size it was bought. See A whole server: bare metal.
Regular compute is the default and is what most sites want. Cores, memory and disk arrive in one plan at one price, on the fastest disk there is. The disk can only get bigger, and more room means moving to a bigger machine.
Cloud native compute sells the disk separately. The machine is cores and memory, and you can make it bigger or smaller whenever you like.
- The disk is a purchase of its own. Choose from 10 GB upwards, in tens to 100 and then in fifties. A machine with a lot of memory and modest files does not need a disk to match.
- It is its own line on the bill. The price follows what the machine's provider charges for its fastest storage. The rate falls as the disk grows; the first ten gigabytes are the dearest.
- The size you choose is the size your site gets. Ten gigabytes bought is ten gigabytes you can fill; the room the platform needs is not taken from it.
- Growing the disk is not a resize. You enlarge the volume and the machine stays the type it is. It still pauses for a moment.
- The resize dialog shows the cost. It lists each size's monthly price and the difference from now. The disk's line on the bill moves to the new size from the day it grew.
- The storage is on the provider's network, not in the machine. The volume is built on the fastest network tier the provider sells.
A whole server: bare metal
Where a datacentre sells them, bare metal is a physical server that is yours alone. Cores, memory and disks are not shared with another customer.
- You see the whole machine. The cores and memory quoted are what is in the server. The platform's own tools (the agent, the proxy in front of your site, backups, logs) can use up to 10–15% of them when busy. Plan for the rest.
- Most servers' disks are in RAID 1. Two disks hold the same data, so one failing does not take your site or your files with it. The list says RAID 1 on each server that has it. One that does not say it keeps your site on a single disk, as a cloud machine does: cheaper, and a failed disk is a restore from the nightly backup.
- Each server says what its disks are. NVMe is the fastest, SSD next. HDD is spinning disks: slow for a database and the cheapest way to keep a great many files.
- The room quoted is what you can fill. It is one disk's, less what the platform keeps beside your files, rounded down to a plain figure. That share holds database dumps and backups before they go offsite, and grows with what you store. It is larger than on a cloud machine: a server's disk cannot be made bigger later.
- Traffic is not metered. Where a server's traffic is not counted, it is listed as unmetered with the speed of its port (500 Mbit/s, 1 Gbit/s) and what that port can carry in a month. That figure is a ceiling, not an allowance: nothing is charged for reaching it and there is no extra bandwidth to buy. The port is shared with other servers, so it is not a promise either.
- A faster port, where the server has one. Many servers can be given a faster port for a monthly price: 1 Gbit/s on the smaller ones, up to 25 Gbit/s on the largest. Choose it when you set the project up, or later under Commercial terms. A faster port is switched on within minutes and nothing restarts; going back to a slower one applies from your next invoice. Some servers are sold with the one port they have; the list says which speeds each can take.
- It is one server and it stays that size. There is no worker or extra machine beside it, no autoscale, and no resize. A bigger server is a new project on a bigger server. Preview environments are built on the usual preview machines, not on it.
- It takes longer to arrive. Allow up to an hour, where a cloud machine takes minutes.
- The mirror fills in after it arrives. A new RAID 1 server is usable straight away, but its two disks are still being made identical in the background. Until that finishes the disks are slower, and only one holds a complete copy. It takes about an hour and a half for 1 TB of SSD, and most of a day for several terabytes of HDD. Nothing needs doing; allow for it before a large import or a load test on a server's first day.
- A reboot takes minutes, not seconds.
- Cancelled, it goes when billing stops. Other machines are kept a day past the end of what you paid for. A bare-metal server is deleted at that moment, with its data. Backups are kept as for any project — see Billing.
It can carry several of your own projects on the shared plan, like any machine that is yours.
Shared or reserved CPU
Sizes are grouped under two headings.
| What it is | Choose it for | |
|---|---|---|
| Shared CPU | Shared with other workloads, and costs less. A busy neighbour on the same hardware can slow it down for a moment. | Most sites, and every staging copy. |
| Reserved CPU | Set aside for this machine alone. | A busy shop, an API with steady traffic, or anything you would notice slowing down. |
Reserved is not a promise of a speed: it says the cores are yours, not how fast the processor under them is.
Which way a machine can move
A machine on cloud native compute moves both ways. Changing the cores and memory never touches its disk. The list shows every type the provider sells in that region, with its CPU, memory and monthly price. A size with less memory than its services need between them (what each service needs) is refused, with the figures.
A machine on regular compute only goes up. Its disk grows with the type, and a disk is never made smaller. Once a machine has moved up, the sizes below it are gone for good. The list only offers the size it is on and larger, and says why.
It is downtime
Everything on the machine is offline for a few minutes when you move a machine on regular compute, or shrink any machine. On a single-machine environment that is the whole site. Plan it like a deploy. This is unlike growing a disk.
The resize is done with the machine off:
- The machine stops.
- Its type changes.
- It starts again.
| Case | What happens |
|---|---|
| Growing a machine on cloud native compute, where the provider allows it | Cores and memory are added while it runs, with nothing offline. |
| Growing it and its disk in the same step | Not while it runs. |
| The provider refuses the new type | The machine is started again at the size it was. |
Before you confirm, the resize dialog says whether the machine stops or keeps running.
A machine resizes with its role. Every machine doing the same job, such as the web servers behind a balancer, is kept at one size. Resizing one resizes the others, one at a time, and the dialog says so first.
Scaling for a while
A production machine on cloud native compute can also be made bigger for a while without changing what it was bought at: by autoscale when it is busy, or by an upscale for seven days. See Scaling.
The disk on regular compute
This section is about regular compute. On cloud native compute the disk is resized separately — see above.
A larger type comes with a larger disk, and nothing ever shrinks it back. So the list stops at the size the machine is on: the smaller types have less disk than the machine already uses.
If you need more room without changing the machine, buy a volume. It grows without downtime and without taking anything away.
Adding a machine for background work
On a dedicated plan, on either shape, you can add a worker: a machine that runs your workers, the long-running processes you declared, instead of the web machine. The schedule stays on the web machine.
Add one when background work and visitors compete for the same CPU. The symptom is a slow site, not a slow job — see Shapes.
You pick its size, and it is charged like any other machine.
What reaches a machine, and what leaves it
The same rules apply on every machine, wherever it runs, over IPv4 and IPv6 alike. They are not a setting.
In from the internet:
| What | Where |
|---|---|
| HTTPS on 443, and HTTP on 80, which redirects to HTTPS | The machine that serves your sites. Behind a load balancer or a cache, the machine in front answers instead. |
| SSH, keys only | The machines your sites run on — see Shell. |
| Ping, and the ICMP that connections rely on, such as "packet too big" | Every machine. |
| Nothing else | A database, cache, search, queue, worker or extra machine answers nothing from the internet. It is reached over the environment's private network. |
Out to the internet, from the machines your sites run on: everything, TCP and UDP to any port, apart from mail.
- Ports 25 and 465 are refused on every machine. The connection fails at once, with "connection refused". Send mail through a relay on 587, which is open everywhere. The OpenSMTPD service does that for a whole environment once it has a relay host.
- The private network is not the internet. A site handing its mail to the environment's own relay on port 25 is not refused.
An extra's way out is its own — see Extras.
Memory
Your environments on a machine may use the memory their plans were sold, and the machine keeps back what it needs to run and to stay reachable. A process that would go past that is stopped by the machine, inside your environments alone: the machine, and every other site on it, carry on. A site stopped this way needs more memory than its plan has, and a bigger machine.
Security updates
Debian's security fixes install themselves once a day on every machine, and nothing else does. A machine never moves to a new release of anything on its own. Most fixes take effect as they install.
A fix to the kernel or the C library takes effect only when the machine restarts. In your project's backup and maintenance window the platform installs whatever is waiting and, where a fix needs it, restarts the machine. Its sites are offline for a minute or two, at the hour you chose for maintenance, and start again on their own.
This is on for every project. Turn it off under Configuration → Backups → Backup / maintenance schedule → Change ("Automatically apply security updates in this window").
A machine that carries several of your projects restarts only when every one of them has it on. Where one has it off, or the project has turned it off, updates still install every day and the machine is marked reboot waiting in the console, beside its Reboot button. Reboot it when its sites can be down for a minute or two.
On a project with more than one machine, one finishes before the next starts.
What you cannot change
The architecture. An arm machine cannot become an x86 one: its disk image will not boot there. Only types of the same architecture are offered.
Next
Thank you. Good to know it helped.
Thanks — tell us what was missing. Open a ticket from Support in the console, and we answer on the ticket. Open the console