[OpenIndiana-discuss] ZFS pool slow as molasses

Brett Dikeman brett.dikeman at gmail.com
Wed Oct 6 14:14:20 UTC 2010


On Wed, Oct 6, 2010 at 4:23 AM, Albert Lee <trisk at opensolaris.org> wrote:
> On Tue, Oct 5, 2010 at 9:11 PM, Chris Mosetick <cmosetick at gmail.com> wrote:
>> If it's not too late already, my suggestion at this point is to go back to
>> ide mode in the bios, boot the machine and rsync your data to another
>> physical machine through the network, then do a clean install of openindiana
>> with your bios set to ahci mode, create new pools, then rsync your data back
>> to the new pools.  Time consuming, yes.  But virtually guaranteed to work.
>>
>> It would seem that this machine would have worked better with ahci mode when
>> you first installed opensolaris on it back in the day.  This is a lesson for
>> everyone to check their bios settings thoroughly before installing a new
>> operating system.
>
> Erm, that's a curious bit of modern folklore. How the drive was
> connected previously changes absolutely nothing.

I agree.  Aside from a few hiccups with having device names changed
(easy to fix), there's nothing permanent done.

> The problem here is probably from having dedup enabled. Disabling it
> does not affect the existing data, which has already been
> deduplicated. Reading any of this will incur a deduplication table
> lookup, which can be quite expensive in terms of random seeks

I've disabled it, and I believe that our backup software's
"maintenance" function re-packs client archives, so within a couple of
days (it's configurable how often this runs per-client), things will
improve, provided the server is able to chew through the archives
quickly enough.

After I enabled dedupe, I never saw the ratio change from 1.00; I
later learned archives are encrypted with a client-specific key, so
theoretically, they're almost completely unique- a worst-case scenario
for the dedup lookup, right?

I'll have more to report in about a week, hopefully good news.  Is
there a way to measure the size of the lookup table?

Sidenote: could someone fix the mailman settings for the list?
They're currently set such that the Reply-To: headers make it annoying
to reply to an individual off-list, destroying the functionality of
reply vs reply-all.

-B



More information about the OpenIndiana-discuss mailing list