Veeam Repository on Tintri IntelliFlash Array

[dpmarkham]

I am working with a customer setting up Veeam B&R for their VMware environment. They have a Tintri IntelliFlash array that they are going to be using for storage for their Veeam Repositories. I have not worked with IntelliFlash arrays previously as a backup target. The Tintri documentation says the following:

Note: Veeam will prompt that your chosen filesystem is not formatted with ReFS. IntelliFlash Repositories should not be formatted as ReFS as NTFS is best practice for SAN volumes. This also provides support for UNMAP, whereas ReFS does not provide this and results in Thin-provisioned LUN’s becoming thick over time (unallocated space not returned).

This documentation is from 2020 but this is the first I have heard of a storage vendor not recommending ReFS for Veeam repos. Seems the advantages provided by Block Cloning with ReFS would out weigh any concerns regarding support for UNMAP. Looking for guidance or anyone with experience with IntelliFlash.

[david.domask]

Heya @dpmarkham,

I'm afraid you're sort of caught between Vendors here.

ReFS doesn't send UNMAP/TRIM commands: https://learn.microsoft.com/en-us/windo ... kup-target

The end result is that deletes from an application to a filesystem that doesn't support these protocols will leave a lot of leftover data without a clear way to clean it.

I'm not sure what Intelliflash does or does not support, but it's best to confirm with them 100% what the requirements are. My past experience is that thick provisioned volumes handled this fine and only thin provisioned had issues with the space reclamation, but I must ask you to really push on the storage vendor as to how they handle XFS/ReFS and the most ideal setup for their storage.

If your client has the space to absorb the full size of the LUN, I'd advise don't think twice about it and just go for it since the block cloning speed benefits should be more than enough to offset the predicted size of the repository. If space is a concern, it's absolutely best to confirm the specific interactions with Intelliflash support. Backup applications (like Veeam) just make a file-system call; the end result is dependent on the hosting storage.

[dpmarkham]

Tintri/Tegile has not been very helpful with support on my customer questions. Keeps just telling them they need to engage Prof Services to do anything. They did send the document regarding configuring the array for Veeam which I pulled the above excerpt from. I was hoping others may have had experience using Intelliflash arrays with Veeam since I have mostly done NetApp. Disk space is not really an issue for them so I may see if they can thick provision the LUNs and then use ReFS. I assume the fact that UNMAP does not work with ReFS does not prevent the repo server from overwriting deleted data on the filesystem. The array may just not reflect the actual used space on the LUN properly but the OS will see it correctly. Is that correct?

[david.domask]

I must admit I'm not a deep storage person, but my understanding was that the lack of UNMAP/TRIM was specifically about reclaiming space on Thin Provisioned LUNs and over-provisioning became an issue; it's what I've seen most often from our Support cases. With Thick-Provisioned, I don't understand this will be an issue as I get it, but take this with a grain of salt as this is not my specialty.

I should note that XFS _does_ handle this reclamation well, so it may be a consideration if they're willing to go the Linux route.