To: Gavin Hamill <gdh@laterooms.com>
cc: pgsql-performance@postgresql.org
Subject: Re: pg 8.1.3, AIX, huge box, painfully slow. 
In-reply-to: <20060407205919.a091da26.gdh@laterooms.com> 
References: <20060407185823.abe93017.gdh@laterooms.com>
	<1144436060.32269.13.camel@state.g2switchworks.com>
	<20060407205919.a091da26.gdh@laterooms.com>
Comments: In-reply-to Gavin Hamill <gdh@laterooms.com>
	message dated "Fri, 07 Apr 2006 20:59:19 +0100"
Date: Fri, 07 Apr 2006 16:06:02 -0400
Message-ID: <15772.1144440362@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Gavin Hamill <gdh@laterooms.com> writes:
> Scott Marlowe <smarlowe@g2switchworks.com> wrote:
>> My guess is that this is an OS issue.  Maybe there are AIX tweaks that
>> will get it up to the same or higher level of performance as your four
>> way xeon.  Maybe there aren't.

> The pSeries isn't much older than our Xeon machine, and I expected the
> performance level to be exemplary out of the box..

I'm fairly surprised too.  One thing I note from your comparison of
settings is that the default WAL sync method is different on the two
operating systems.  If the query load is update-heavy then it would be
very worth your while to experiment with the sync method.  However,
if the bottleneck is pure-SELECT transactions then WAL sync should not
be a factor at all.

Does AIX have anything comparable to oprofile or dtrace?  It'd be
interesting to try to monitor things at that level and see what we can
learn.  Failing a low-level profiler, there should at least be something
comparable to strace --- you should try watching some of the backends
with strace and see what their behavior is when the performance goes
south.  Lots of delaying select()s or semop()s would be a red flag.

			regards, tom lane