quick zfs question hopefully - RAIDZ or RAIDZ2?

farscapesg1

2[H]4U
2FA
Joined
Aug 4, 2004
Messages
2,658
So, I'm almost ready to move everything off my old WHS server over to my new OI+Nappit server and I'm trying to decide the best option for my drives for data storage (movies, music, etc.). I have 5 WD20EADS drives and 3 Hitachi 7200 RPM drives to put into an array. My original thoughts were to do 4 WD20EADS in a RAIDZ config + the 3 Hitachi in a second RAIDZ stripe, with the last WD20EADS as a hotspare. I had orginally planned on adding my old 1.5TB green drives as a third RAIDZ stripe for extra storage, but after getting a used DroboPro free from work I want to use them in that for a backup of critical files.

Would I be better off mixing the drives into a single RAIDZ2 + hotspare for 10TB usable space? Forget the hotspare and put all 8 in RAIDZ2 for 12TB space? Or go with my original plan? My current WHS only has about 5TB of total space, so it's not like I "need" all 12TB usable space...
 
I would go with a single vdev of raid2z. 2 drives can fail before data is comprimised. If this is for media, then preformance is not a large factor.
 
I'm an N+2 person as you can lose 1 of something and still have redundancy...i.e. go with RAID-Z2!
 
RAIDZ2.
When one disk fails on RAIDZ1, you have no back-up left while you're trying to rebuild the array. In the meantime, extra stress is put on the remaining disks during access and/or resilvering to recreate the missing data based on parity, increasing exponentially the likelihood of a second disk crash.
Resilvering could last several hours BTW if your storage counts several TB.
 
I seem to recall that with raidz*, all disks need to be read for a single block (unlike standard non-zfs raid), so resilvering shouldn't change the stress on remaining disks (no more so that a regular scrub, IMO...)
 
OK, thanks for al, the input :) Looks like the consensus is RAIDZ2. I'll probably do a 7 disk vdev keeping the last disk as a spare. 10TB available space should be enough storage space for now :)

I guess mix and matching drive types and spindle speed isn't an issue...
 
There really isn't a lot of point in hotspares unless you're at the highest redundancy level.

I would rather use RAIDZ3 than RAIDZ2+hotspare.
If a drive fails in a RAIDZ3 setup, you're basically running a RAIDZ2, but in case a drive fails in the RAIDZ2+hotspare array, you're running RAIDZ1 AND stressing your drives with a resilver.

Most likely, though, you'd be fine running an 8 disk vdev RAIDZ2 :)


edit: Above suggestion is only in the case of you only using 1 vdev for your pool. My understanding is that hotspares are pool wide. So obviously, for more vdevs, the switch to RAIDZ3 instead of RAIDZ2 would require a drive per vdev, but going with RAIDZ2 the hotspare(s) would just get used by the vdev in which a drive fails.
 
I seem to recall that with raidz*, all disks need to be read for a single block (unlike standard non-zfs raid), so resilvering shouldn't change the stress on remaining disks (no more so that a regular scrub, IMO...)

You're right on that one, but resilvering is stressful because of the raw amount of I/O in short timeframe used to recreate the array (like a scrub, yes).
 
Back
Top