3x SSDs as a stripe and 1x large HDD as a mirror

That basic shoud work. A colleague use luks under zfs also but said in this way of luks per disk the overhead is higher than with “native”.

1 Like

Edit: Upps, I should read carefully as you mentioned partitioning the hdd in 3 parts … but which hdd is 3 times as big as 1 of your ssd’s or otherwise there is some space lost on hdd or ssd side.

The idea is 3x 8TB SSD and 1x 24TB HDD (split into 3 partitions).

It’s going to, for the most part, perform like two mechanical hard drives. Not like three SSDs. If that’s a consumer hard drive, I’d also be concerned about premature wear due to all the constant and I mean CONSTANT frantic seeking across partitions (the partitions themselves aren’t the important part, it’s that the partitions force the drive to write in rapid succession to WIDELY physically separated areas of the media as it tries to keep up with those three SSDs).

2 Likes

Be aware, reading may not be fast as ZFS doesn’t guarantee it will try to read from the SSD’s but will sometimes read from the rust disk. So you will randomly get slow, high-latency reads as well as fast ones. Mixing SSD’s and rust disks in a mirror is generally not a good idea.

I’ve put a script here that is how I replicate from a local SSD and a local rust disk to a couple of other (also local) rust disks:
https://gitlab.com/jaytuck/linux-settings/-/snippets/3738526

rpool is the SSD.
slowdisk is the rust disk.
slowdiskBFG and slowdisk9VS are the backup disks.

I’ve also put the systemd files to run it on a schedule.

The script also saves out a “status file” that you can then monitor for A) is it recently modified and B) is the contents “0” and if so you know that syncoid is running as expected.

2 Likes