Re: postgresql.conf - Mailing list pgsql-performance

From scott.marlowe
Subject Re: postgresql.conf
Date
Msg-id Pine.LNX.4.33.0307301403110.25368-100000@css120.ihs.com
Whole thread Raw
In response to Re: postgresql.conf  (Josh Berkus <josh@agliodbs.com>)
Responses Re: postgresql.conf  (Josh Berkus <josh@agliodbs.com>)
Re: postgresql.conf  (cafweb <cafweb@lulu.com>)
List pgsql-performance
On Wed, 30 Jul 2003, Josh Berkus wrote:

> Ron,
>
> > I don't know if this is representative of other postgresql installs, but
> > I would also put in my vote for the differentiation added, as these are
> > not small machines but are multi-server boxes.
>
> But how is the Multi-purpose configuration different from the Small Machine
> configuration?    If the actual settings are the same, we just need to
> explain somewhere what it means.
>
> I'll argue pretty strongly against including a seperate MP configuration
> because it would raise our number of suggested sets to 10 from 7.

Maybe we should look at it more from the point of view of how much
horsepower (I/O bandwidth, memory, memory bandwidth, cpu bandwidth) is
left over for postgresql.  After all, a Dual 2.8GHz Opteron with 32 gigs
of ram is gonna be faster, even if it has apache/LDAP/etc  on it than a
dedicated P100 with 64 meg of ram.

I think the default postgresql.conf should be the one for the 64 Meg free
PII-300 and below class, and our first step up should assume say, 256 Meg
ram and simple RAID1, approximately 1GHz CPU or less.  The high end should
assume Dual CPUs of 1Ghz or better, 1Gig of ram (or more).

Once someone is getting into the 8 way Itanium II with 32 Gigs of RAM,
the fact that they are doing something that big means that by looking at
the default, the workgroup, and the large server configs, they can
extrapolate and experiment to determine the best settings, and are going
to need to anyway to get it right.

So, maybe just a note on which parameters to increase if you have more
RAM/CPU/I/O bandwidth in the big server example?


pgsql-performance by date:

Previous
From: Rajesh Kumar Mallah
Date:
Subject: Re: Why performance improvement on converting subselect
Next
From: Josh Berkus
Date:
Subject: Re: postgresql.conf