Re: Setting Shared-Buffers

2G per process is plenty ...and useful if you have large data warehouse style queries which are long running (especially multiple of those)

We do benefit from the Linux memory caching model regardless of what Postgres uses right ?

On a machine which we upgraded from 4G to 16G on a 32 bit PAE kernel...we saw a doubling of performance for most queries of a certain type.(mostly data warehouse type accessing several hundreds of thousands of records).

Postgres version that we use is 8.1.9.

-----Original Message-----
From: pgsql-admin-owner(at)postgresql(dot)org on behalf of Scott Marlowe
Sent: Sat 7/11/2009 3:04 AM
To: Anj Adu
Cc: Tino Schwarze; pgsql-admin(at)postgresql(dot)org
Subject: Re: [ADMIN] Setting Shared-Buffers

On Fri, Jul 10, 2009 at 3:26 PM, Anj Adu<fotographs(at)gmail(dot)com> wrote:
> You can use upto 64G of RAM on a 32 bit RHEL 5/ Fedora 8 OS using the kernel
> PAE extension.

And it's about 15% slower, and pgsql itself can only access ~2 or 3G
shared and 2G per process. I routinely set shared_buffers to well
over 3G on big machines, and have a few reporting queries that run
truly huge work_mem settings. Really, there's not much reason to be
running postgresql on 32 bit unix anymore, unless you're stuck using
an ancient flavor or something.

However, I was referring to Windows, where things are even worse, as
the OS only sees 3Gigs total cause apparently it doesn't support PAE.