(N+P) Stripe Width Reccomendation benchmarks?

[zeroluck]

I have seen several references to this "Rule" about RaidZ stripe width. The "Rule" goes as follows:

(n^2+p) drives for RAIDZ sets, where n is a number and P is the RAIDZ level.
OR
(N+P) with P = 1 (raidz), 2 (raidz2), or 3 (raidz3) and N equals 2, 4, or 8
So,

RAIDZ 3, 5, or 9 drives
RAIDZ2 4, 6, or 10 drives
RAIDZ3 5, 7, 11 drives

Sources:

https://forums.freenas.org/index.php?threads/getting-the-most-out-of-zfs-pools.16/#post-52
https://forums.freenas.org/index.php?threads/getting-the-most-out-of-zfs-pools.16/#post-2043
http://www.solarisinternals.com/wik...onfiguration_Requirements_and_Recommendations
https://forums.freenas.org/index.php?threads/raid-z1-4-drives-bad.6066/#post-22273
https://forums.freenas.org/index.php?threads/raid-z1-4-drives-bad.6066/#post-22216

You'll notice that link is dead, but I found a PDF version of it from 2010 here: https://documents.irf.se/get_document.php?group=Computer&docid=311. If you scroll to page 6, you'll see where this idea comes from, and a link of where whoever wrote this now-missing document (maybe) got the idea, but that link is also broken. If you do more internet detective work, you'll find a copy of that blog post (from 2008) here: https://blogs.oracle.com/timthomas/entry/recipe_for_a_zfs_raid

There is also dissent about this:

https://forums.freenas.org/index.php?threads/raid-z1-4-drives-bad.6066/#post-22275

My question is: Has anyone actually done benchmarks on this or tested it at all to see if there is any real world truth to it? I see no mention of it in the ZFS docs from Oracle. I also haven't seen any actual math backing it up.

 

[dlavigne]

Our ZFS primer states this:

• Some older ZFS documentation recommends that a certain number of disks is needed for each type of RAIDZ in order to achieve optimal performance. On systems using LZ4 compression, which is the default for FreeNAS® 9.2.1 and higher, this is no longer true. See ZFS RAIDZ stripe width, or: How I Learned to Stop Worrying and Love RAIDZ for details.

Note that the referenced URL was written by Matt Ahrens, one of the original ZFS designers and the main developer for OpenZFS. He wrote it specifically after asking me "why do you say that" after a presentation I gave that mentioned the old rules. I asked if I could get the "official" answer in writing and he wrote that post as the response.