pg.ddx.io  pgsql-performance@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: Aaron Krowne <akrowne@vt.edu>
To: Sean Chittenden <sean@chittenden.org>
Cc: pgsql-performance@postgresql.org
Subject: Re: postgresql meltdown on PlanetMath.org
Date: Sun, 16 Mar 2003 03:06:01 -0500
Message-ID: <20030316080601.GF19570@vt.edu> (raw)
In-Reply-To: <20030316061208.GA62529@perrin.int.nxad.com>
References: <20030316060125.GD19570@vt.edu>
	<20030316061208.GA62529@perrin.int.nxad.com>

> Have you read the following?
> http://developer.postgresql.org/docs/postgres/performance-tips.html

Yup.  I would never go and bother real people without first checking the
manual, but I bet you get a lot of that =)

> This shouldn't be an issue for the load you describe.  A p-100 should
> be okay, but it depends on your queries that you're performing.

Mostly just gather-retrieval based on unique identifier keys in a bunch
of tables.  Really mundane stuff.

> *) Stick with btree's.

Yeah, that saddens me, though =)  When I initially switched to hashes,
things were blazing.  This application makes heavy use of keys and equal
comparisons on indices, so hashes are really the optimal index
structure.  I'd like to be able to go back to using them some day... if
not for the concurrency issue, which seems like it should be fixable
(even having mutually exclusive locking on the entire index would
probably be fine for this application and would prevent deadlock).

> > - I then tried increasing the machines shared memory max to 75% of the
> >   physical memory, and scaled postgresql's buffers accordingly.  This
> *) Don't do this, go back to near default levels.  I bet this is
>  hurting your setup.
> > - Lastly, I tried reducing the shared memory max and limiting postgresql
> >   to more conservative values, although still not to the out-of-box
> >   values.  Right now shared memory max on the system is 128mb,
> >   postgres's shared buffers are at 64mb, sort_mem is at 16mb, and
> >   effective cache size is at 10mb.
> *) You shouldn't have to do this either.

Well, I've now been advised that the best way is all 3 that I have tried
(among aggressive buffering, moderate buffering, and default
conservative buffering).  

Perhaps you could explain to me why the system shouldn't be ok with the
moderate set of buffer sizes on a 512mb machine?  I don't really know
enough about the internals of postgres to be doing anything but voodoo
when I change the values. 

> I hate to say this, but this sounds like a config error.  :-/

Thats better than a hardware error!  This is what I wanted to hear =)

> *) Have you done a vacuum analyze?

See previous message to list (summary: it worked this time, but usually
it does not help.)

Thanks,

Aaron Krowne



view thread (39+ messages)  latest in thread

Message-ID: <20030316080601.GF19570@vt.edu>
Permalink:  ../20030316080601.GF19570@vt.edu/
Also on:    postgresql.org/message-id/20030316080601.GF19570@vt.edu

 · 

reply

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 from your mail client; the link fills in
  the headers below: reply via email
  Some clients, webmail especially, drop In-Reply-To and References from
  a link, and the reply then starts a new thread: use the mbox below.

  To: pgsql-performance@postgresql.org
  Cc: akrowne@vt.edu, sean@chittenden.org
  Subject: Re: postgresql meltdown on PlanetMath.org
  In-Reply-To: <20030316080601.GF19570@vt.edu>
  References: <20030316060125.GD19570@vt.edu> <20030316061208.GA62529@perrin.int.nxad.com> <20030316080601.GF19570@vt.edu>

* 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