You could, but:
If you really want, you could run the HDDs in a mirror VDEV as well as a metadata special VDEV with 2 NVMe in a mirror.
But really, don’t use USB flash for SCALE’s boot drive.
All ZFS pools have a ZIL, ZFS Intent Log, used for synchronous writes. However, an external one is know as a SLOG / LOG, Separate intent LOG. As Dan said, most SOHO users don’t need a SLOG. Let alone Mirrored SLOGs that Enterprise users may want.
L2ARC devices can be striped but not Mirrored. And again as Dan said, you don’t really have enough memory for using a L2ARC. Each entry in the L2ARC requires RAM for pointers / directory, so reduced RAM with a L2ARC, just reduces RAM further.
Don’t use a USB thumb / flash drive. However, either a SATA SSD or NVMe in an USB enclosure can work. It is just that many USB thumb / flash drives are of exceptionally poor quality that one might last just months. That said, their are some USB drives that are of higher quality and cost is noticeably higher as a result.
I ordered one on AliExpress. It came with a 16-GB RAM module and 512 GB NVMe drive. I upgraded it with a 32-GB DRAM module and ordered two 16-TB drives from Go Hard Drive.
AOOSTAR R1 (16 GB/512 GB)—$254.37
32GB DDR4 3200MHz PC4-25600—$52.99
2 × Seagate Exos (factory re-certified)—$349.90
16 TB of redundant storage for $657.26.
My impression so far is that installation of TrueNAS worked like a cinch. The hard drive caddies are not the best, but they work. I still have to burn in the drives.
I think, this will work as a backup solution for my residential clients, although I’ll source the NAS through Amazon, instead.
Just a follow-up. It looks like this solution works well, well enough, at least, for a 2-drive NAS. I have not done much with it, yet. Installation of TrueNAS was straightforward. I have burned in the drives (each took about eight days to burn in with the disk-burnin-dak180.sh script by @dak180).
I will try using it in a client’s office, for backing up his Windows workstations.
n00b question… why is there a need to burn-in these drives? From what I’m reading “burn-in” is basically testing the drives (load/stress testing) to make sure if there’s a defect they fail earlier rather than later (when say the warranty has (in)conveniently expired)? Would this burn-in procedure be more valuable for when you have used/re-furbished drives rather than brand new ones?
Ideally you burn-in test every drive you plan to add, no matter how you sourced it.
New drives can fail, hence the popularly referenced “bathtub curve”, as can used ones (that you don’t know the history of).
Nope. Infant mortality is definitely a thing; I’ve had more than one drive fail soon after putting it into service. Better to catch it before you have data on it.
I burnt-in new drives, but when I bought refurbished data centre drives to replace some of the batch I put the drives straight into the pool, as I trusted that 30kh of previous good work (and passing SMART tests) were enough of a burn-in…
Drives are particularly prone to failing after they’ve been moved.
The movement from the factory to into your system is normally not kind.
Best to find any early failures doing a surface test than when you’ve nearly filled it with data.
The reason is called the “Bathtub curve”.
The detailed reson is that there are certain manucaturing defects that cannot be catched by a (usually really short) End Of Line testing that is done in the HDD factory (Testing costs a LOT! fully testing and burning in a drive would make them even much more expensive. It is unfrotunately much cheaper for the manufacturer to skip that step and outsorce it to the end user. At least much cheaper than dealing with the RMAd drives later))
And Etorix is right: Refurbished drives are already “Burnt in”, so there is no need to do this to them.
