Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 22 Nov 2009 21:36:18 -0500
From:      jhell <jhell@DataIX.net>
To:        Randy Bush <randy@psg.com>
Cc:        freebsd-fs@freebsd.org
Subject:   Re[2]: 7.2 dies in zfs
Message-ID:  <917839072.20091122213618@DataIX.net>
In-Reply-To: <m2my2efiko.wl%randy@psg.com>
References:  <m2my2hkj2y.wl%randy@psg.com> <20091122154228.GB55532@rron.freenix.org> <m2my2efiko.wl%randy@psg.com>

next in thread | previous in thread | raw e-mail | index | archive | help
=0D=0ASunday, November 22, 2009, 2:47:03 PM, Randy wrote:

>> I've given up on ZFS on i386.  Whatever tuning you could do is only
>> delaying the inevitable.  Even with lots of RAM, it will panic.  I'd
>> love being proven wrong as I also hav a 4 GB i386 with ZFS and it
>> panics regularely.

> i am not sure one can not tune it to stay up.  it's just not clear how.
> and that is the critical point.

> on a 4g i386 with moderate+ load, my current parms are
>     vm.kmem_size=3D1500M
>     vm.kmem_size_max=3D2G
>     vfs.zfs.arc_min=3D120M
>     vfs.zfs.arc_max=3D900M
>     vfs.zfs.prefetch_disable=3D1
> and it's up over two days, wooo wooo!

> i suspect that i will next lower arc max.

7.2-STABLE ---
arc_min & arc_max AFAIR are auto tuned on boot by values that dont
come to mind right now. Have you tried just leaving these for the
system to tune....

On my i386 running 7.2-STABLE with 1G RAM 1800MHz with two 7200RPM
drives I dual boot with Win7 & FreeBSD not that it makes a difference.
My large drive carries FreeBSD, ZFS only with two slices of my second
drive one for the ZFS cache and one for the ZIL.

kmem_size is 512M and max is 768M & I dont disable the prefetch...

Kernel config invludes KVA_PAGES=3D512 along with no other kernel
tuning.

Seems no matter what I do with this machine I can not get it to crash
whatsoever. I have stress tested it with unixbench, iozone & dd(1).

This machines uptime during long periods of not having to boot into
(Windows for Work) purposes is around 13 days or so. ;-)

I could not be more pleased other than I would like some faster write
speeds when doing inbound xfers from a personal fileserver but this I
just blame on lack of better hardware.

Best of Luck.

PS: If you go too low on arc_* you are going to inhibit your write
speeds dramaticly while trying to save your memory for other purposes.
--=20

 Sunday, November 22, 2009 9:18:42 PM

 jhell




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?917839072.20091122213618>