ZFS on top of a bunch of Raid0 (1 disk) volumes with battery backed write-cache

Question

I'm reinstalling a workstation (not a headless server) with an Intel RAID card with BBU and 4GB internal cache.

My current setup uses ZFS on a pair of mechanical disks and a partition on an SSD for the ZIL.

I want to keep using ZFS on 2 bigger disks, mainly because storage flexibility: filesystems and block devices. That comes handy for lxc containers, ability to make regular snapshots, etc.

The controller i'm about to use doesn't support write-caching on JBOD drives, but it does on 1 disk volumes with RAID0. That's the closest it gets to JBOD without loosing the battery backed write cache.

The main concern I have is the lack of SMART information available to ZFS. I can query the drives directly with smartctl -d megaraid,S /dev/sdb, being S the drive slot on the controller. I'm not sure if ZFS if capable of using that.

As I said, this is for a personal workstation. What do you think? Having thousands of fsyncs/s thanks to the write cache, do I need ZIL o can I live without it?

Any specific settings on the controller (regarding the SMART management) that I should take into account?

Thanks

Answer 1

It can be done of course, but getting some recent and decent HBA instead is a waaaaay more bulletproof route.

Answer 2

ZFS does not use SMART values in any way. Not sure where you got the idea that it does.

Answer 3

Please take a look at any introduction to ZFS. It is imperative that ZFS uses HBA controllers, not RAID controllers (with cache). ZFS needs full and direct control of disk access. This includes no write caching. I repeat, imperative. All may look fine, but the failure will come at the moment you least want it: rebuilding a failed array, etc.

Even if this is for testing only, you will not be learning ZFS properly.

Performance improvement on ZFS comes with using high speed ARC and SLOG devices, not write caching.