[QUESTION] How to distribute 8 * 4TB disks optimaly on zpools/vdevs?

[Dmitry Nevozhay]

My system is E3C226D2I mobo, Xeon E3-1245v3, 16GB ECC RAM, LSI 9211-8I in IT mode, 8-bay U-NAS NSC-800 case, 400W PSU, 8 * HDD WD40EFRX RED 4TB.
FreeNAS 9.2.1.6.

It should mainly store and stream (NFS, DLNA, plex) ripped collection of media (lot of FLACs, blu ray isos), and store my personal file archive (raw photos, documents).
Loss of ripped media is not fatal (but not desired).
Loss of personal files is fatal...
Number of simultaneously connected media clients is 2 or may be 3 (mediaplayer via NFS, tv with plex plugin, ipad with itunes).
The amount of memory installed (16GB) is already maximum allowed my my mobo.
(I would glad to increase it, but it's impossible)

I would build the single zpool of 1 vdev of 8 disks in raidz2... is it a good idea?
But as I've read, the recommendation for raidz2 vdev is 6 disks...
So another option I consider is 2 zpools - 1st one for file archive on 2 disks (mirror),
and 2nd one for media library on 6 disks (raidz2).
Or 2 zpools, both of raidz1 - of 3 and 5 disks respectively...

What do you think, guys, what option to choose?

 

[rs225]

Loss of personal files is fatal...

I would go with 1 drive for your off-site backup, and the other 7 in a raidz2. Or, if you're concerned about the optimal raidz2, go with 6 in raidz2, and have 2 drives for off-site backup(s).

 

[danb35]

The "optimal RAIDZ vdev size" issue has become, with the advent of compression, a non-issue--feel free to create your vdevs with whatever number of disks you desire (I'd probably just go for an 8-disk RAIDZ2, once the disks were thoroughly tested). A motherboard with only two DIMM sockets was probably not the best choice for FreeNAS, but 16GB will likely be adequate for your needs. Be aware, though, that Plex is very RAM-hungry.

 

[sjieke]

I'm looking into the setup of an identical NAS (8 drives) with similar use cases. Following is a blog post I found which is an interesting read (http://blog.delphix.com/matt/2014/06/06/zfs-stripe-width/). I don't know how correct the numbers are, but according to this you should have at least 6 drives for raidz2, anything above that should be fine.
A simple qoute

A misunderstanding of this overhead, has caused some people to recommend using “(2^n)+p” disks, where p is the number of parity “disks” (i.e. 2 for RAIDZ-2), and n is an integer. These people would claim that for example, a 9-wide (2^3+1) RAIDZ1 is better than 8-wide or 10-wide. This is not generally true.

A question to the experts: can we believe what is written in this article, our is it rubish?

 

[danb35]

The author of that blog post is one of the lead ZFS developers, so it's safe to say it isn't rubbish--that post is the main basis for my post. There may be a remaining question of how applicable it is to any given use case--it seems to deal mainly with database use, involving relatively small record sizes being written to disk, which may not correspond well to (say) media storage, which has fewer, and much larger, files being stored.