If you have never lost a file system, I'm guessing you haven't had very many.
Also, ZFS gracefully handles cheap disk. RAID cards don't handle the case where you are using consumer grade disk and it retries for hours rather than failing. (I recently tested this on a rather nice 3ware 9550SX. It handled the problem no better than linux md. Performance was worse on the 3ware, too. I'm sending it back.)
I am personally switching to 'enterprise' (read, 20% more cost for 30% less capacity) sata just so I get drives that actually fail rather than retrying when they are failed.
And this is ignoring the 'silent corruption' issue the CERN guys were on about. Block level checksumming fixes that too.
Also, cheap snapshots are awesome. LVM snapshots work, and they can save your ass in a pinch, but they are expensive performance-wise. In my case, where I'm pushing my disk to it's performance limits already, LVM snapshots are nearly unusable. ZFS snapshots, I understand, have almost zero performance overhead. (space is cheap. Performance, not so much.)
This has nothing to do with ZFS, but FYI: you can configure consumer-grade Western Digital drives to report failures immediately, rather than retrying, just like their more expensive "enterprise" counterparts. See http://en.wikipedia.org/wiki/Time-Limited_Error_Recovery for details. A bit of Googling will turn up the WDTLER.exe binary that you need; make a bootable FreeDOS CD or USB stick, then run WDTLER from the command line to toggle the feature on or off.
Possibly there are other good reasons to buy the enterprise models, but using the cheaper drives is working out for me. I'm running 6 WD15EADS drives with TLER disabled in a software RAID6 configuration on a GNU/Linux machine, and except for the fact that two of the drives failed within hours of first use (the replacement drives work fine), I've had no problems running the array 24/7 for over a month now.
Yeah, I have 2 wd 'black' consumer grade drives fixed with WDTLER paired with some seagate ES drives right now. (I mix makes to avoid firmware plagues) I read that WDTLER only works on 1TB and smaller drives. I will have to find some 1.5TB WD drives and experiment.
As for the relation to ZFS, it is supposed to timeout ill-behaving consumer drives like this. In fact, from what I understand, you can configure ZFS's behavior, for example, to fail the first drive that times out, and then hang on the second drive (which would perhaps be a reasonable configuration for a ZRAID volume)
If you have never lost a file system, I'm guessing you haven't had very many.
Also, ZFS gracefully handles cheap disk. RAID cards don't handle the case where you are using consumer grade disk and it retries for hours rather than failing. (I recently tested this on a rather nice 3ware 9550SX. It handled the problem no better than linux md. Performance was worse on the 3ware, too. I'm sending it back.)
I am personally switching to 'enterprise' (read, 20% more cost for 30% less capacity) sata just so I get drives that actually fail rather than retrying when they are failed.
And this is ignoring the 'silent corruption' issue the CERN guys were on about. Block level checksumming fixes that too.
Also, cheap snapshots are awesome. LVM snapshots work, and they can save your ass in a pinch, but they are expensive performance-wise. In my case, where I'm pushing my disk to it's performance limits already, LVM snapshots are nearly unusable. ZFS snapshots, I understand, have almost zero performance overhead. (space is cheap. Performance, not so much.)