[rescue] The State of SunHELP (new server build, etc)

[Bob Darlington]

zfs raidz specifically eliminates the write-hole issues.  On the smallest
of my systems I'm running 20 8TB disks raidz3 with a hot spare, times 4 for
84 disks in the jbod, and typically 4 of these craptastic seagate jbods to
a server.  I can't vouch for smaller systems, but I will say my stuff is
fairly quick.  Here's some of the settings...

History for 'stg01':
2016-02-22.15:15:34 zpool create -o ashift=12 stg01 raidz3 /dev/sdb
/dev/sdc /dev/sdd /dev/sde /dev/sdf /dev/sdg /dev/sdh /dev/sdi /dev/sdj
/dev/sdk /dev/sdl /dev/sdm /dev/sdn /dev/sdo /dev/sdp /dev/sdq /dev/sdr
/dev/sds /dev/sdt /dev/sdu spare /dev/sdv -f
2016-02-22.15:39:59 zpool export -a
2016-02-22.15:44:57 zpool import -d /dev/disk/by-id stg01
2016-02-29.13:48:31 zfs set compression=lz4 stg01
2016-06-17.14:50:36 zfs set reservation=20G stg01
2016-10-21.10:20:31 zfs set recordsize=8M stg01
2016-10-21.10:24:14 zfs set relatime=on stg01

pretty stock stuff.   Watch the -f's  (force).

-Bob

On Mon, Jun 10, 2019 at 2:14 PM Jonathan Patschke <jp at celestrion.net> wrote:

> On Mon, 10 Jun 2019, Bob Darlington wrote:
>
> > Raidz3 all the way.   There are other under the hood tweaks we use and
> I'll
> > get you the list when I am in the office and remember.
>
> Benchmark your load before going with raidz3.  In order to avoid the
> parity "write hole", the I/Os will tend to serialize, which means you'll
> usually get throughput limited to the slowest disk in the batch.
>
> For the same reason, tuning recordsize on raidz can make a huge difference
> (partial stripe writes with parity make everyone have a rough day).
>
> I've had excellent experience with doing a zpool where each vdev is a
> mirror and devices in the mirrors are split between manufacturers (ie:
> WDC in the "left" side of each mirror, and Seagate in the "right").
>
> For my "back up the world" box, it's raidz2 over slow disks.  It works
> fine, but performance is pretty bad for modern hardware.  For the machine
> I run data-crunching jobs on, I have two zpools each with four disks
> arranged into two mirrors.
>
> For additional tweaks and tunables, please refer to the quote in my
> signature. :)
>
> --
> Jonathan Patschke   |   "The more you mess with it, the more you're
> Austin, TX          |    going to *have* to mess with it."
> USA                 |                            --Gearhead Proverb
> _______________________________________________
> rescue list - http://www.sunhelp.org/mailman/listinfo/rescue