To: "Luke Lonergan" <llonergan@greenplum.com>
cc: "Gavin Hamill" <gdh@laterooms.com>, pgsql-performance@postgresql.org
Subject: Re: pg 8.1.3, AIX, huge box, painfully slow. 
In-reply-to: <C05C3088.21195%llonergan@greenplum.com> 
References: <C05C3088.21195%llonergan@greenplum.com>
Comments: In-reply-to "Luke Lonergan" <llonergan@greenplum.com>
	message dated "Fri, 07 Apr 2006 15:06:48 -0700"
Date: Fri, 07 Apr 2006 18:12:39 -0400
Message-ID: <17237.1144447959@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Luke Lonergan" <llonergan@greenplum.com> writes:
> On 4/7/06 3:02 PM, "Tom Lane" <tgl@sss.pgh.pa.us> wrote:
>> On the other hand, we already know that Xeons suck about as badly as
>> can be on that same measure; could the pSeries really be worse?

> I wouldn't be too surprised, but it sounds like it needs a test.  Do we have
> a test for this?  Is there a contention-prone query stream that we can think
> up?

If you want you could install a pre-8.1 PG and then try one of the
queries that we were using as test cases a year ago for spinlock
investigations.  I don't recall details right now but I remember
having posted a pretty trivial test case that would send a
multiprocessor machine into context-swap storm, which sounds a whole
lot like what Gavin is seeing.

I think that 8.1 ought to be relatively free of buffer-manager spinlock
contention, which is why I doubt that test case would be interesting
against 8.1.  The interesting question is what else is he seeing
contention for, if it's not the BufMgrLock?

			regards, tom lane