I currently have an NVME device as cache/pool device that hosts my VM and also my appdata / docker. I use mover to move download files to the array once per day. The disk is formatte as XFS. It works flawless and there is per se no need for change.

I feel like tinkering a bit and making some further improvement. I am thinking to buy a second NVME to mirror / backup my current cache device. Assume this should be quite easy. A few questions:

• Would this work with my XFS drive or would I need to use ZFS or BFRS for this?

• If so, is ZFS or BRFS suggested?

• Is conversion as easy as copying the files from the XFS disk to another ZFS disk?

Thanks for your help!

3 hours ago, steve1977 said:

Would this work with my XFS drive or would I need to use ZFS or BFRS for this?

Only ZFS or Btrfs support multiple device pools.

3 hours ago, steve1977 said:

If so, is ZFS or BRFS suggested?

Recommend ZFS.

3 hours ago, steve1977 said:

Is conversion as easy as copying the files from the XFS disk to another ZFS disk?

Basically, yes, you can also use the mover.

Thanks for your quick response. Let me go ahead with ZFS then. A few for follow-ups that I get it right:

1) Mover indeed may be the easiest. I would need to stop docker and VM before doing so? Worried that some r/w takes place during "mover" activity.

2) I'd like to mirror the NVME disk(s). Anything I'd need to keep it mind or just set it up as a RAIDZ1 pool?

3) I only have one NVME disk now, but I will buy another one. Needs to be same size, but can be a different model/speed?

4) Eventually, I may upgrade to a total of 4 or even 6 NVME disks (in total). Besides the joy of tinkering, is there any benefit to separate appdata/docker, download/cache, and VM files to separate disks? Right now, I have all three of it on one NVME (currently XFS without mirror, to be changed to ZFS with mirror). If I were to go to 4 or 6 NVME disks some time in the future, would I set them all up as ZFS? Anything I should keep in mind when setting up the first pool now?

17 hours ago, steve1977 said:

1) Mover indeed may be the easiest. I would need to stop docker and VM before doing so? Worried that some r/w takes place during "mover" activity.

https://docs.unraid.net/unraid-os/using-unraid-to/manage-storage/cache-pools/#moving-files-between-a-pool-and-the-array

17 hours ago, steve1977 said:

2) I'd like to mirror the NVME disk(s). Anything I'd need to keep it mind or just set it up as a RAIDZ1 pool?

Default for a new two device pool is a mirror

17 hours ago, steve1977 said:

3) I only have one NVME disk now, but I will buy another one. Needs to be same size, but can be a different model/speed?

Yes, can also be larger.

17 hours ago, steve1977 said:

4) Eventually, I may upgrade to a total of 4 or even 6 NVME disks (in total). Besides the joy of tinkering, is there any benefit to separate appdata/docker, download/cache, and VM files to separate disks? Right now, I have all three of it on one NVME (currently XFS without mirror, to be changed to ZFS with mirror). If I were to go to 4 or 6 NVME disks some time in the future, would I set them all up as ZFS? Anything I should keep in mind when setting up the first pool now?

That's more of a preference/use case, you can have a single pool with more drives, better for performance, or separate pools.

Thanks @JorgeB .

re 1 - Link explains to move from cache to array. In my case, I would use the mover to move from pool (XFS disk) to a newly created one-disk pool (ZFS disk). And then I reformat the XFS to ZFS and add it as second disk to the ZFS pool?

re 3 - If larger, it would only use the the size of the smallest disk though?

re 4 - Let me simplify my question. I have more than sufficient space on my NVME disk for docker, VM and scratch folder (for arr downloads). Would I be better off to separate these three things (i.e., dedicated disk for VM, dedicated disk for dockers, dedicated disk for scratch folder for downloads). Or does it not really matter and I can keep all three on one NVME disk?

And another question - I just read that I could also format one of my array disks as ZFS and then mirror the cache disk to the array. This way, I wouldn't need a second NVME disk and the cache disk would even be double-protected (mirrored and protected by parity). Is this a smart thing to do? I am wondering whether I'd lose a lot of performance as the NVME would suddenly be much slower as it woudl be mirrored to a regular sata hdd? Also not sure whether this would slow down the VM. I remember running the VM on an array before, which was super slow. So, if I now run the VM on an nvme cache disk, but mirror it to a sata array disk, would this be the same and slow?

8 hours ago, steve1977 said:

I would use the mover to move from pool (XFS disk) to a newly created one-disk pool (ZFS disk)

It's the same; just select the other pool instead of the array.

8 hours ago, steve1977 said:

If larger, it would only use the the size of the smallest disk though?

Yep

8 hours ago, steve1977 said:

I just read that I could also format one of my array disks as ZFS and then mirror the cache disk to the array.

Not clear if you think it's mirrored in real time; it's not. You would need to have a script or something doing snapshots and then send/receive, but it could also be a suitable option, again if you don't mind it's not real time.

Got it. Your suspicion was right. I indeed thought yould create a (real-time) pool between cache zfs nvme disk and array zfs sata disk. In fact, I thought I read this somewhere. If that's not the case, then the answer is that I will add two nvme disk to keep them mirrored. Assume the mirroring doesn't have a massive impact on performance?

It shouldn't, and with ZFS reads can be faster from a mirror since the data is striped.

Thanks again for your help. The transition from the old XFS to the new ZFS disk worked well using the mover approach. I re-enabled VM and Docker. However, there is one issue now, which feels quite odd.

I only have on GPU. I forgot what I had assigned it to prior to the change. However, It is now assigned to both my Plex docker container and also my VM. I always use both my VM and my Docker at the same time. So, this would not have been possible with this assignment. Unless I found a way to assign it to both in parallel?

I have now just removed it from my VM, but weirdly enough getting an error message when doing so.

Any idea why this has happened? Shouldn't the docker and VM settings be unchanged after disabling / enabling them? Or was there some way how to assign the GPU to both and still run them in parallel?

On 9/15/2025 at 3:37 PM, JorgeB said:

On 9/14/2025 at 9:48 PM, steve1977 said:

1) Mover indeed may be the easiest. I would need to stop docker and VM before doing so? Worried that some r/w takes place during "mover" activity.

https://docs.unraid.net/unraid-os/using-unraid-to/manage-storage/cache-pools/#moving-files-between-a-pool-and-the-array