mix ReFS & XFS extent in Scale Out repository

[M4rco]

I would like to better understand why mixing ReFS & XFS repositories in a Scale Out repository is a bad idea.

veeam-backup-replication-f2/refs-vs-xfs ... ml#p388907

Here it is stated that there are many reasons not to mix but the most important is the difference in blocksize.

I would expect the following 2 settings to be of influence:

* per-machine backup files
* data locality

If both these settings are enabled than I don't understand why a difference in blocksize would matter. A VM would end up on either the ReFS extent or the XFS extent and just stays there.

I am also interested to hear about the many other reasons.

Being able to mix ReFS & XFS extent would make the transition from ReFS to XFS a lot easier!

[Andreas Neufert]

Hi M4rco,

you are right the block size on the file system or hardware has no effect on the Veeam processing. Our block size is always bound to the job and do not matter where the data lands in the end. You can mix both together but think about the following:

1) I guess it is way more common scenario to not mix technologies and SOBR. So more customers use this configuration and I guess it is more well tested within Veeam QA. => Likely more stable processing.
2) Veeam has commented to work on features like data migration between extends without breaking the block cloning. I don´t know about the plans and timing but I guess this would be available within same technologies.

[M4rco]

Thank you for your confirmation.

As I understand there is really no reason (other than: this setup has probably been tested less by Veeam QA) why I cannot or should not mix repositories with different filesystems in a SOBR.

[Gostev]

Correct. SOBR was specifically designed for mixing storage technologies together, as our goal was to enable customers to use whatever storage devices they already have more efficiently. This was literally one of the primary use cases! Because of that, no SOBR logic requires or assumes matching storage technologies within a Performance Tier... it's really the opposite.

[M4rco]

Even stronger confirmation! Thank you.

Please note: Veeam support (Case # 04608051) stated this:

"From my point of you it would be better to have a homogeneous infrastructure and do not mix XFS and REFS in the same SOBR.

It might work until the data of the backup chain stays localy on one of the file systems. If the chain would create some of the increments on the extent with another filesystem due to unavailable free space or if the primary extent is offline there might be issues, at least with the fast clone."

What issues can I expect when data locality can no longer be upheld?

[Gostev]

I have no idea what are they talking about. Fast clone does not work across SOBR extents in any case! Thus, having a "homogeneous infrastructure" makes zero difference. If a backup file cannot be created on the "home" extent, then it does not really matter what extent it spills on to. It's not like having all-ReFS or all-XFS SOBR will allow fast clone to continue in the two situations they presented.