Backup fo VMs slow responsde and big

[drnicolas]

Every day a backup job runs.
It contains 9 VMs, but i see funny effects:
- running VMs becomes almost irresponsive during the backup; before and after everything is fine.
- 1 VM needs almost 4h to backup; the size is alwaysaround 790G. the interesting fact: this VM is swiztched off since weeks.

The backup goes to a NFS storage

 

[Dunuin]

1 VM needs almost 4h to backup; the size is alwaysaround 790G. the interesting fact: this VM is swiztched off since weeks.

Not sure how dirty-bitmapping then would behave. My guess would be that only running VMs can make use of dirty-bitmapping (because without QEMU running the vdisks could be manipulated without QEMU being able to notice it) so the whole 790GB would need to be read and hashed even if nothing has changed. So it wouldn't send anything to the PBS as nothing had changed but the hashing still needs some time to check if anything had changed.

- running VMs becomes almost irresponsive during the backup; before and after everything is fine.

Everything is done client-side (reading data, hashing data, compression, optional encryption) so this causes a lot of CPU load and IO. Use "fio" to benchmark your PVEs storage and use the "proxmox-backup-client benchmark" command to benchmark your CPU performance and "iperf3" to benchmark your network performance to see where the bottleneck is.

The backup goes to a NFS storage

PBS will be slow over network shares or when using HDDs. Local SSD storage is highly recommended.