To: Christopher Petrilli <petrilli@gmail.com>
Cc: Vivek Khera <vivek@khera.org>, pgsql-performance@postgresql.org
Subject: Re: Impact of checkpoint_segments under continual load conditions 
In-reply-to: <59d991c405071907482bb0689b@mail.gmail.com> 
References: <59d991c4050717100874c08a86@mail.gmail.com>
	<A50B8A1F-71A1-4F63-A5F4-1AA58F89557D@khera.org>
	<59d991c405071811457ccdd3f1@mail.gmail.com>
	<16394.1121715156@sss.pgh.pa.us>
	<59d991c405071812342e290f57@mail.gmail.com>
	<16794.1121718743@sss.pgh.pa.us>
	<59d991c405071907482bb0689b@mail.gmail.com>
Comments: In-reply-to Christopher Petrilli <petrilli@gmail.com>
	message dated "Tue, 19 Jul 2005 10:48:42 -0400"
Date: Tue, 19 Jul 2005 11:04:30 -0400
Message-ID: <24014.1121785470@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Christopher Petrilli <petrilli@gmail.com> writes:
> Here's a graph of the differences and density of behavior:

> http://blog.amber.org/diagrams/pgsql_copy_803_cvs.png

> I can provide the raw data.

How about the complete test case?  There's something awfully odd going
on there, and I'd like to find out what.

> Note that fsync is turned off here.  Maybe it'd be more stable with it
> turned on?

Hard to say.  I was about to ask if you'd experimented with altering
configuration parameters such as shared_buffers or checkpoint_segments
to see if you can move the point of onset of slowdown.  I'm thinking
the behavioral change might be associated with running out of free
buffers or some such.  (Are you running these tests under a freshly-
started postmaster, or one that's been busy for awhile?)

			regards, tom lane