I’ll start by saying I am not the expert when it comes to ZFS, and that I cosplay as a sysad with my Homelab. I’m a data science guy, so this is at the edge of my expertise.
This is my general build plan, hopefully folks around here can weigh in and warn me of any pitfalls:
My current NAS is a TrueNAS Scale box running 1 vdev w/ 6x 4tb HDDs in RaidZ1. It’s in an old NZXT H440 case at the moment.
I use it to serve a postgresql DB (on spinners… I know… hence the upgrade), nextcloud, lan SMB shares, and (of course) Jellyfin. I don’t need it do anything but NAS things.
I’ve budgeted out an upgrade path, as I expect my data needs to grow in the next few months.
- Upgrade to an HL15, ASRock x570 Creator, a pcie x16 nvme x4 riser card (1st x16 slot), and an LSI pcie x8 HBA w/ 4x SFF-8643 Card 3rd x16 slot).
Migrate my data to a temporary store, move the drives over (is there anyway to keep the pool with a move like this, or will it have to be rebuilt?) Setup Houston UI to run the NAS, migrate the data into place.
3x10tb drives in rz1 w/ 2tb L2Arc & 2tb slog nvme drives - this will handle most of my data as I migrate things around
5x SSD’s in rz2 for databases (should this have l2arc or slog drives? Do those need to be individual drives or can 1 drive have the slog for 2 pools? All my reading leads me to believe that having an slog drive for a Database Pool greatly improves performance)
eventually replace the 4tb drives with bigger ones. (as it goes)
I’d like to have 2 nvme drives in Raid0 as a working drive for models and training data, and rsync that to more stable storage regularly.
There will be on-site and off-site backups maintained on other systems.
Does this seem like a reasonable plan? I’d like to do step 1 now, and the next steps every month or two after. Will I run into any issues with this plan? Have I setup something silly? Am I a fool for trusting in the speedy sorcery that is nVME storage and not going for 15x 20TB Seagate Exos as our Rust-Powered diety demands?
Thanks in advance for any comments or advice! I only got into homelabbing a few months ago, and I’m excited to build something that I can run production level datasets on.