Since I started experimenting with ZFS geometry and the like today, I think it’s finally time to do the rust storage inventory as well:
Rust storage inventory (2024-02-12)
I think I’ve re-tested and validated all the remaining HDDs in my possession. There should still be at least one 4TB drive in my old-old machine that I will recover one day, but for now I’ll count it as a spare.
I initially ordered more 3TB and 4TB drives because they had very good $/TB ratio. Unfortunately half of the order turned out to be DOA and I had to document and return them. Given the time I spent on that, if I had a choice I’d just go with refurb 12TB X12s/X14s, but I digress.
Here’s what’s available:
1x ST5000LM000 (5TB Barracuda Compute, 2.5") - I currently use it as primary data backup and don’t plan to use it in the pool (but likely modernize filesystem and data layout to work better with new ZFS datasets)
2x ST33000650SS (3TB Constellation ES.2, SAS) - 2 out of 4 arrived DOA so I’m not holding my breath here. I’ll likely use them as secondary, low-use backup. The 2 that arrived working seem fine for now, but are quite old nevertheless.
4x HGST HDN724040ALE640 (4TB Deskstar NAS) - good track record with these so far.
1x WDC WD4000F9YZ (4TB WD Se) - 1 out of 2 arrived DOA. The sample size was small but I wouldn’t trust it too much.
2x ST1000DM010 (1TB Barracuda Compute) - OK. Not good, not terrible. Above average transfer rates.
2x WDC WD10S21X (1TB WD Black SSHD, 2.5") - The drives are quite funny. They have 8G SSD R/W buffer that makes some workloads super fast. Not sure how they’ll fit here, but included nevertheless.
1x HGST HTS541010A7E630 (1TB Travelstar, 2.5") - I don’t even know where I have this one from.
1x WDC WD10JPCX (1TB WD Blue, 2.5") - Likewise
The last 3 positions (4 disks) are the most egregious, but they can work pretty well in tandem. Since they are all 1TB drives I have tested grouping them in md raid0 and they readily beat any other drive on the list not only in speed, but also in power usage - as primarily mobile parts they are exceptionally good at it.
I wouldn’t trust them too much, however, and will only use them as either a spare, or as a +1 in a RAID5 (RAIDZ1).
Summing by size we’ve got:
1x5TB likely off-limits
5x4TB
2x3TB
6x1TB (4x LP SFF, 2x LFF)
ZFS tank proposal
Excluding special devices (log, cache and special) I currently envision the pool to be either 5x4TB RAIDZ1 + 1x4TB spare or 6x4TB RAIDZ2. The 6th 4TB to be constructed as md raid0 from the SFF 1TB drives. This leaves:
1x5TB
2x3TB
2x1TB
For hot backups or “extra-spares”. Why extra spares? Because if needed I can temporarily convert 1x3TB + 1x1TB into “4TB” using LVM. I even researched using LVM for a more permanent solution, but I couldn’t find a way to get asymmetric stripes between the drives (i.e. have 3 stripes on the 3TB drive for each stripe on the 1TB drive) and forewent that route.
A 4x1TB matrix can be easily created with:
sudo mdadm --create /dev/md0 --verbose --homehost=any --level=0 --raid-devices=4 --name=md4x1Ta /dev/sd[cdjl]