[OpenIndiana-discuss] Very slow write performance on a 3 disk raidz1-0 zpool

[Robbie Crash]

Just disabling dedup isn't going to help, as ZFS is lazy and won't rewrite
the deduped data, so you'll still have your dedup tables paging. You need
to disable it, and rewrite all the data in a way that doesn't just update
the link. For example, moving the data to a different directory on the same
FS won't do anything, but moving it to a different FS will.

Honestly though, it will take you so long, and be so annoying, and so
frustrating, that you're better off copying everything to USB and nuking
the pool then starting from scratch.

I had to go through the same thing with a 1.4 dedup ratio on an 80% full 10
TB pool on a box with 32GB a few months back. Moving 8TB at 5MB/sec, not so
much.

Sended from my Android.
On Aug 24, 2012 7:46 AM, "Bryan Iotti" <ironsides.medvet at gmail.com> wrote:

> Everything I posted about revolved around the impression that you had
> terrible performance because of the incorrectly reported block size (many
> newer drives do this, it does affect performance in some cases but not this
> bad).
>
> Sašo is perfectly right, dedup on a system with 1GB RAM is going to grind
> it to a halt.
>
> Disable dedup on the pool, or if you want to do things optimally you could
> copy the data off the pool, recreate it with only compression and recopy it
> over...
>
> Zpool-12 was mentioned because the system-supplied one only creates pools
> with an ASHIFT value of 9. This means that on a 512 byte per block disk the
> pool starts properly on a block boundary, while on a 4k per block Advanced
> Format drive it starts mid-block (hope I got this right, holler if I'm
> wrong) and the whole pool is skewed, with the disk controller working in XP
> compatibility mode and yielding poor performance. For this reason someone
> created zpool-12, a modified version of the zpool command that allows you
> to specify the ASHIFT value and avoid this issue. Still won't prevent the
> freakin' firmware on the Caviar Greens from spinning them down very often,
> though!
>
> Bryan
>  On Aug 24, 2012 12:31 PM, "Julius Roberts" <hooliowobbits at gmail.com>
> wrote:
>
> > On 24 August 2012 18:02, Bryan Iotti <ironsides.medvet at gmail.com> wrote:
> > > Are these Advanced Format Drives (4k sectors) already?
> >
> > I don't know?
> >
> > > My own WD Caviar Green 2TB sucked bad before I recreated a pool on it
> > with
> > > the modified zpool-12 command.
> >
> > This pool was created in 2008(?) using zfs v10 on solaris express
> > developer edition, and was recently upgraded to v5000 or whatever it
> > is that comes with OIa5.
> >
> > > You could run:
> > > # zdb <pool name> | grep ashift
> > >
> > > And see what value you get. If you have 9, the pool is structured for
> 512
> > > byte sectors and a 4k drive could suffer a performance loss. For a 4k
> > > drive, the ASHIFT value should be 12.
> >
> > twsadmin at darling:~/mdbackup/today$ sudo zdb Backups | grep ashift
> > Password:
> >                 ashift: 9
> >                 ashift: 9
> > ^C
> >
> > What does this mean exactly and what is the expected performance loss
> > in this situation?
> >
> > --
> > Kind regards, Jules
> >
> > <golgy> whats so wrong with plumb?
> > <hoolio> nothing, in itself.  it's just for me, knowing what it means
> > infers i cannot any longer pretend to not be a complete square when it
> > comes to computers
> > <Gryphon> I don't know that knowing anything about plumb turns you
> > into a nerd, but this conversation already has
> > <hoolio> are you calling me nerdy?
> > <checkers> hoolio: you know what initramfs means, AND does. You're
> > lost to the non-geek world already
> > <Gryphon> yes
> > <hoolio> hrm
> > <hoolio> goodbye cruel world.
> >
> > _______________________________________________
> > OpenIndiana-discuss mailing list
> > OpenIndiana-discuss at openindiana.org
> > http://openindiana.org/mailman/listinfo/openindiana-discuss
> >
> _______________________________________________
> OpenIndiana-discuss mailing list
> OpenIndiana-discuss at openindiana.org
> http://openindiana.org/mailman/listinfo/openindiana-discuss
>