REFS - Backup methods - Tape

[futureweb]

Hi,

Earlier we used:
- Forever Incremental backup, once a week the tape job made a "virtual Full Backup".

The problem was, that the tape job was cancelled once a normal backup job started. As tape jobs take a long time, this happened quite often.

We changed to:
- Incremental backups with weekly Active Full Backups (Took twice as disc space). The Tape job didn't have troubles anymore.

From support I got told, the reason is, that files get locked and the tape has to cancel the job.


Now I would like to know if this behavior changed with REFS and we could switch back to forever incremental.
As if I understand correct, the "virtual Full Backup" only links to blocks from the initial full backup, and wouldn't lock anything.

Can you confirm, as we would like to move to REFS to have less disk usage and decrease the active full counts.

Also, is there a best practices yet for REFS, other than using 64K?

Thanks
Patrick

[DaveWatkins]

We run reverse incremental because of this, but there is also a new tickbox in tape jobs to tell them not to release the backup files if a job wants access, this means while you may miss a backup your tape jobs will complete successfully. It's under Schedule and called something like Prevent this job being interrupted by the primary backup job

[futureweb]

Yes, I found this option, but I was wondering, if it's still necessary with REFS, but otherwise, I think of following:

- Incremental Backups Synthetic Full once a week (usually tape should take this as full and not doing Virtual Full if I remember right)
- and have this option enabled, to wait.

How do reverse incremental work with REFS and performance. We never did it because of I/O Overhead pre 9.5

thanks
Patrick

[DaveWatkins]

It's hard to say, I converted from RAID10 to RAID6 when I converted to ReFS so the speed benefits we're mitigated by that. I'd still say synthetic operations are faster now than they were on the RAID10 with NTFS but because of all the changes I don't know for sure (also upped the stripe size when I rebuilt the arrays).

We initially went reverse incremental before any of this was an option to remove the need to build the virtual full to push to tape. We might be able to switch back now without issue however I like the idea of being able to remove old vbr files to get space back if it ever comes to that, since with forward incremental your oldest file is the large VBK and deleting it breaks the chain, however with reverse incremental deleting the oldest vbr file just removes your oldest backup and all the rest are fine