A backup of your VPS every night, with fourteen days of history. Think about that for a second and a question comes up: fourteen copies of, say, an 80 GB disk is more than a terabyte per customer. Fortunately, that is not how it works.
The answer lies in incremental backups, and in the way Proxmox Backup Server handles them. That is the software we run the backups of our VPS hosting on.
Full, incremental, and the hassle in between
There are roughly two ways to make a backup:
- A full backup copies everything, every single time. Simple and reliable, but slow and extremely expensive in storage if you want it daily.
- An incremental backup copies only what changed since the previous run. Fast and economical, but classically fragile: to restore, you need the full backup plus every incremental step after it. If one link in that chain is missing, the rest is worthless too.
That is why you used to run, say, a weekly full backup with daily incrementals in between. Workable, but you are left with a chain that can break and with one heavy moment every week.
How Proxmox Backup Server solves that
Proxmox Backup Server takes a different approach. Instead of thinking in files or whole disks, it cuts your data into pieces: chunks. Every chunk gets a SHA-256 checksum as its name, and that checksum decides whether it already exists.
If a chunk does not change, its checksum is identical and it is simply not stored again. That is called deduplication, and it has a pleasant side effect: every backup is internally a complete list of all the chunks your system consisted of at that moment. It is just that most of those chunks are already there from earlier runs.
The result is the best of both worlds:
- You only send the changes, so a daily backup is small and fast.
- Every restore point is a complete restore point. You do not need a chain of previous backups; you can restore any snapshot on its own.
And because identical chunks are stored only once, a series of fourteen daily backups takes nowhere near fourteen times the space.
The dirty bitmap: why it is so fast
Sending only the changes is one thing, but how does the system know which blocks changed without reading the entire disk first?
For that, Proxmox uses a dirty bitmap. As long as your VPS is running, the virtualisation layer keeps track of which blocks on the virtual disk are being written to. At the next backup the system therefore knows straight away which blocks to read, without touching the rest. That saves an enormous amount of disk I/O, and it is exactly why a nightly backup barely burdens your server.
One nuance: that bitmap only exists while the VM is running. Restart or shut down your VPS and the bitmap is gone, so the next backup has to read the whole disk again to work out what changed. That one backup takes longer than usual. It is not a fault, and afterwards things simply return to the fast method.
What else is in there
Proxmox Backup Server does more than copy cleverly:
- Zstandard compression, which handles several gigabytes per second and so never becomes a bottleneck.
- Client-side AES-256-GCM encryption, where data is already encrypted before it reaches the backup server.
- Verification. The checksums make it possible to check afterwards whether stored chunks are still intact, rather than just assuming they are.
- Prune and garbage collection. A retention schedule decides which snapshots stay; chunks no longer referenced by any snapshot are cleaned up afterwards.
- Remote sync. A backup server can forward its data to another location, again sending only the differences.
On top of that it is simply open source, released under the AGPL v3, so you can check exactly what it does.
What that means for your VPS
In concrete terms: a backup every day, fourteen days of history, and each of those fourteen points is a complete restore moment. You never have to wonder whether there is still a full backup underneath and whether the chain is intact.
And you do not have to ask anyone. In our VPS panel you see the available restore points, pick the moment you want to go back to, and start the restore yourself. No ticket, no waiting, not even at three in the morning.
Run into something, or unsure which restore point you need? We are of course happy to think along through our Discord or a ticket.
Want to know more about how backups relate to storage, and why the two are not the same? That is covered in our article on backup and storage.