Intel Optane P905P on Unraid - Need Recommendations

[werkkrew]

Hi all -

I am in the process of building a new server for my Unraid setup which will include two 1.6TB Intel Optane 905P's.

Worth noting I will be essentially moving my existing array setup onto new hardware so the base array will stay as-is.

For the array as it sits I am using the standard Unraid array with the disks formatted as XFS.

For the Unraid "cache" that I use in the general sense of it being a write cache primarily for my downloads and such. This is two 1TB Samsung 990 Pro's configured using BTRFS "raid1" as an Unraid cache pool.

I am not sure how best to configure the Optane. I spent a lot of money on them so it would be nice to be able to realize their full performance benefits. My intention is to use them for my docker/appdata/vm disks. Ideally they will be protected with some form of software raid (zfs or btrfs raid1 I guess) but I am afraid that formatting them in this manner the filesystem overhead might negate the benefit of the drives performance. Is this valid?

I would consider running them in a non-raid protected layout if it was worth it performance wise and make up for the lack of raid with increased frequency of my backups if needed.

 

[nexox]

Depending on your workload, a CoW filesystem like ZFS or BTRFS may not be as fast as you would like. Realistically any layer on top of Optane is going to add some latency, but you probably want some kind of filesystem, so compromises have to be made. I personally run mine in lvm and configure certain logical volumes that require redundancy as mirrors, but mdraid is also an option, in either case XFS is probably the fastest filesystem to run on a volume. I also don't put VM disks on a filesystem as an image, I run them directly from logical volumes, not sure how useful that is but it shaves off a layer.

 

[werkkrew]

Depending on your workload, a CoW filesystem like ZFS or BTRFS may not be as fast as you would like. Realistically any layer on top of Optane is going to add some latency, but you probably want some kind of filesystem, so compromises have to be made. I personally run mine in lvm and configure certain logical volumes that require redundancy as mirrors, but mdraid is also an option, in either case XFS is probably the fastest filesystem to run on a volume. I also don't put VM disks on a filesystem as an image, I run them directly from logical volumes, not sure how useful that is but it shaves off a layer.

So in the context of Unraid, can I do this?

 

[nexox]

So in the context of Unraid, can I do this?

I have no experience with Unraid, can't help with the specifics there.

 

[name stolen]

Personally, and this is truly just me, in a home or homelab type scenario, I would skip redundancy on Optane and use the full capacity. Yes, the drives can occasionally have some weird random error or even fail, but anything can, and with Optane you won't be using up the write endurance, so it won't fail that way.

 

[werkkrew]

Personally, and this is truly just me, in a home or homelab type scenario, I would skip redundancy on Optane and use the full capacity. Yes, the drives can occasionally have some weird random error or even fail, but anything can, and with Optane you won't be using up the write endurance, so it won't fail that way.

That is what I was leaning towards honestly, if I increase the frequency of my backups I can mitigate the risks for the most part.

I still don't think Unraid will allow me to make a pool with anything other than ZFS or BTRFS, unless I use the unassigned devices plugin or format the disks outside of the unraid gui and mount them at the OS level manually - which is fine except I'd feel better working within the confines of unraid so that my configuration is supported.