Any solution to slow backups?

[tangerine]

We are currently running a 8 node cluster - 5 nodes are for computing, 3 nodes are for ceph. It's all in HA on a 10GB network - backups to NFS. Our data transfer speeds are around 40MB/s which is unacceptable for our needs. Curious if migration our backup solution to PBS improves this problem? From research it seems like QEMU and 64k blocks is the bottle neck.

Thanks in advance

 

[dietmar]

From research it seems like QEMU and 64k blocks is the bottle neck.

I assume your VMs use ceph? If so, qemu use 4MB block size for backup (not 64 kb).

Maybe the NFS target is slow? Trty to backup to a local storage and see if that helps.