Date: Tue, 20 Apr 2010 10:39:36 -0700
From: David Kerr <dmk@mr-paradox.net>
To: pgsql-performance@postgresql.org
Subject: Very high effective_cache_size == worse performance?
Message-ID: <20100420173936.GA50886@mr-paradox.net>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.4.2.3i

Howdy all,

I've got a huge server running just postgres. It's got 48 cores and 256GB of ram. Redhat 5.4, Postgres 8.3.9.
64bit OS. No users currently.

I've got a J2EE app that loads data into the DB, it's got logic behind it so it's not a simple bulk load, so
i don't think we can use copy.

Based on the tuning guides, it set my effective_cache_size to 128GB (1/2 the available memory) on the box.

When I ran my load, it took aproximately 15 hours to do load 20 million records. I thought this was odd because
on a much smaller machine I was able to do that same amount of records in 6 hours.

My initial thought was hardware issues so we got sar, vmstat, etc all running on the box and they didn't give
any indication that we had resource issues.

So I decided to just make the 2 PG config files look the same. (the only change was dropping effective_cache_size 
from 128GB to 2GB).

Now the large box performs the same as the smaller box. (which is fine).

incidentally, both tests were starting from a blank database.

Is this expected? 

Thanks!

Dave