Single-NVMe in a Pool (Stripe) for Home Use

[mroptman]

What are the pros/cons of running a NVMe based pool with a single drive?
Anyone experience issues with SSD/NVMe based pools?
The stripe NVMe pool would have frequent snapshots replicated to a raid-z2 pool in case the single NVMe device (in the stripe pool) fails.

NVMe Workload:
- VMs (10 or fewer)
- Container Data
- Home use, not commercial (can restore from replicated snapshots if outage happens)

Running a comparable workload on ESXi with a single NVMe drive and there have been zero issues for over 7 years. Backup strategy with ESXi free is not as robust as with SCALE's ZFS capability w/snapshots and replication. On ESXi, the VMs are shutdown and exported to .7z archives. The intent is to automate this w/ZFS by moving to SCALE (once stable).

All in all, it seems like a waste of NVMe space to run a SSD-based mirror when SSDs are already so reliable. A mirror could be done but wasting 50% space again seems 'wasteful' for non-commercial workload. Curious for others input on their usage of NVMe/SSD pools.

 

[morganL]

As long as you can cope with loss of data between the outage time and last snapshot, it's a very reasonable strategy and keeps the costs down.

The annual failure rate of good SSDs is well below 1% for the first few years.

 

[ChrisRJ]

I have been doing the same thing (conceptually) with XCP-ng and FreeNAS 11.3U5 for a while. My scenario and reasoning are identical. It works quite well and I even replicate the weekly backups to cloud storage (my upstream bandwidth is only 12 Mbps).

 

[Arwen]

Can't say anything about VMs or container usage, but I have my media server data striped. It's actually striped across most of a mSATA 1TB SSD & most of a 2.5" 2TB SATA laptop hard drive. Started doing that about 2015. Since it's cheaper media, (laptop hard drive...), I do get pool errors every now and then. Because it's only a file or 2, easy restore from backups. (Thank you ZFS for telling me WHAT I lost!) And of course, the restore is both on line and non-fatal.

Note that the OS pool IS Mirrored, from a piece on each storage device. Mostly so that I don't have to bother with the more complicated boot from alternate media, then perform the restore. Plus, the OS takes up a trivial amount of space compared to the media.

 

[drinking12many]

I had a single NVME pool for a bit, mostly non-critical VMs for work testing, after awhile I added another and to create a mirror, long as you know the risks etc no reason not too.

 

[sretalla]

The real risk is expiry of the snapshot(s) before you may detect any file corruption that can arrive but not be corrected due to the stripe (and no parity or second copy to correct from).

If you've already accepted the potential for lost data, maybe not a big deal, but it's a second risk which may arrive without total drive failure.