Backup/Snapshots Questions

[Just1n]

Hello All,

I have setup a Server with PVE 8.3.0 on it and have some questions about how the backups work, I am running 5x Linux Consoles with an average of 200Gbs each. If I snapshot or backup them what would be the amount of data that would be saved as a duplicate, how much would be compressed. If I did a full stop backup on all 5 is there a good way to estimate time on each and the data amount it would gain.

Thanks,
Justin

 

[_gabriel]

VM's snapshots are not included in VM backup.

Dedup is done with Proxmox Backup Server, where same data cannot be written twice.

 

[powersupport]

Hi,

When you back up a VM in Proxmox, it only saves the data that’s actually used, not the empty space. The backup is also compressed, so the file size is usually smaller than the VM’s disk size.

For example, if each VM is 200 GB but only 100 GB is used, the backup might be around 60–100 GB depending on how well the data compresses.

A “stop” backup (when the VM is turned off) is usually faster and more consistent. The total backup time depends on your disk speed, compression type, and network speed if you’re backing up to another server.

If you use Proxmox Backup Server, only the changed data is saved after the first backup, so future backups will be much smaller and faster.

 

[Onslow]

A “stop” backup (when the VM is turned off) is usually faster [...]

FWIW, I observe the opposite .

Here you are my observations (posted in https://forum.proxmox.com/threads/p...ather-than-just-used-space.165057/post-813185 ):

"usually subsequent backups of running VMs, executed in "snapshot" mode, last substantially shorter. At least backups to PBS (I don't observe this speedup when creating backups to local storage).
This speedup is thanks to "dirty-bitmap" because of it the backup process knows which data have changed since the previous backup.
So, paradoxically, backup of a running VM can be much faster than of a stopped one.

I'll give an example from my server.
Creating a subsequent daily backup of a running VM takes 2 minutes.
Creating a subsequent daily backup of the stopped clone of the same VM takes more than 5 minutes. Despite of the fact, that the data in the running VM is changing all the time and the stopped clone is stopped for many days, so in the summary this is clearly written:
"reused xxxx GiB (100%). And all "write: " entries are "0 B/s" - because there were no writes. But the process had to read all the data, having no dirty-bitmap to help it."