CBT not working with Veeam Agent on servers?

[TimWalker]

Hi,

I've been using Veeam B&R in production for several years backing up our virtual servers, we use Hyper-V 2019. We are running Veeam B&R 11

We recently had an extended power outage on one of our buildings, and have also been contemplating a move across to Proxmox, so with a couple of our usual hypervisors offline we transferred a few of our VM backups across to Agent backups so that we could perform backups while the servers were running on their replica hosts, and could continue to backup if migrated to proxmox or back to their original hypervisors.

Jobs were created through the 'new agent backup job' wizard on the console in managed by backup server mode, there are four servers running windows server 2012 r2 and one running server 2016.

It seems that the Agent jobs are not using Changed Block Tracking, or not using it very well in any case as the backups are taking rather longer than they were as VM jobs and relatively huge amounts of data are being read for the jobs (so perhaps 20x as much is being read as actually transferred, sometimes more - with our VM jobs this is more like 1.5x).

Is this just the way it is with agent jobs or is something going wrong? It's still finishing safely within our backup windows so not a huge issue.

I've been running the Veeam agent in free configuration on my desktop and laptop for a while and they seem to be working on a more normal 2x read to transferred multipler (one of those is windows 10, the other windows 8.1).

[Dima P.]

Hello Tim,

Mind me asking if you have CBT driver installed for youк Windows hosts or change block tracking is performed without CBT driver? Thanks!

[TimWalker]

CBT driver wasn't installed, the docs give the impression that the Veeam CBT driver isn't necessary for our workloads (NTFS, nothing particularly database heavy).

I have installed it to the protection group now and will update in a day or two with how it has effected things.

[Dima P.]

I'd say CBT driver makes incremental backup more precise in terms of detecting the changes, but the main benefit should be in job processing time. Please let us know how it goes and do not forget to share the backup size / backup time stats. Thanks!

[TimWalker]

So yes, CBT driver seems to be working well, read data to transferred multiplier is around 2x, same as I was seeing with my client PCs on the free agent, and only a little worse than the VM jobs (which are closer the 1.5x multiplier).

Duration much quicker than I was seeing. Only small caveat is I noticed today that one of the servers which was rebooted for Windows updates last night has done a complete read, presumably as CBT wasn't able to track changes during the reboot, pretty manageable once a month for us anyway.

Thanks!

[Gostev]

Normally reboots shouldn't affect CBT driver's ability to track changes, so something else happened that made the CBT driver restart change tracking for safety reasons. If this keeps repeating, I'd strongly suggest that you investigate it with support, as something is off and you never know what these corner cases can lead to. Thanks!