The same author did some more testing recently, also quite interesting to read.
Articles like this always make wonder what the zfs user base ratio is of home users to business users. Running raidz2 at home to eek out more storage efficiency is surely always the right thing to do when number of disks is >=6.
I’m using RAID-Z1 at home with three disks in a NAS. With 4 disks, I’d have 1/3 more cost and only marginally better fault tolerance (both handle a single disk failure, mirroring handles two disks failing if they’re in different VDEVs but not if they’re in different ones). I mostly use the NAS over WiFi, so the performance doesn’t matter that much and it’s not sufficiently critical that spending a day or two restoring it from backups would be more than a minor annoyance.
If I had performance constraints that affected some business outcome, I’d probably do it differently but at that point I’d probably be thinking about HAST and iSCSI as well as ZFS.
I run multiple mirror vdevs at home, as my plan was to move my image library from my current setup to my NAS. I just wanted some more IOPS hence going with mirror vdevs.
I’d disagree with the conclusion. The facts are without dispute. Multiple narrow RaidZ2 vdevs, or even RaidZ3, make sense over mirroring simply for the case that two drives can fail, and you don’t want those two drives to fail in the same vdev.
Consider restoring from your backups. If that is an acceptable option, then just use a z2 or z3 and if you want to upgrade capacity, take down your pool, make a new one, and restore from backup (or even just have two pools while you transition, if you have the headers for it).
If it is not acceptable to restore from backup, then you can’t be in a position where two disks can fail and leave you scrambling to restore.