From: Kevin Barnard <kbarnard@speedfc.com>
To: pgsql-performance@postgresql.org
Subject: Re: Increasing number of PG connections.
Date: Tue, 03 Feb 2004 09:34:01 -0600
Message-ID: <401F6B09.32118.342102@localhost> (raw)
In-Reply-To: <Pine.LNX.4.33.0402021631120.20659-100000@css120.ihs.com>
References: <401E8493.30458.332D5CEF@localhost>
On 2 Feb 2004 at 16:45, scott.marlowe wrote:
> Do you have the cache set to write back or write through? Write through
> can be a performance killer. But I don't think your RAID is the problem,
> it looks to me like postgresql is doing a lot of I/O. When you run top,
> do the postgresql processes show a lot of D status? That's usually waiting
> on I/O
>
Actually I'm not sure. It's setup with the factory defaults from IBM. Actually when I
start hitting the limit I was surprised to find only a few D status indicators. Most of the
processes where sleeping.
> what you want to do is get the machine to a point where the kernel cache
> is about twice the size or larger, than the shared_buffers. I'd start at
> 10000 shared buffers and 4096 sort mem and see what happens. If you've
> still got >2 gig kernel cache at that point, then increase both a bit (2x
> or so) and see how much kernel cache you've got. If your kernel cache
> stays above 1Gig, and the machine is running faster, you're doing pretty
> good.
>
I've set shared to 10000 and sort to 4096. I just have to wait until the afternoon
before I see system load start to max out. Thanks for the tips I'm crossing my
fingers.
--
Kevin Barnard
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: kbarnard@speedfc.com
Subject: Re: Increasing number of PG connections.
In-Reply-To: <401F6B09.32118.342102@localhost>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox