Re: FW: Tx forecast improving harware capabilities. - Mailing list pgsql-performance

From David Hodgkinson
Subject Re: FW: Tx forecast improving harware capabilities.
Date
Msg-id 701BBF0B-1E25-4BC0-BF7C-0D53EAD2784D@hodgkinson.org
Whole thread Raw
In response to FW: Tx forecast improving harware capabilities.  ("Sebastian Lallana" <slallana@datatransfer.com.ar>)
List pgsql-performance
On 18 Aug 2005, at 16:01, Sebastian Lallana wrote:


> It exists something like this? Does anybody has experience about
> this subject?

I've just been through this with a client with both a badly tuned Pg and
an application being less than optimal.

First, find a benchmark. Just something you can hold on to. For us, it
was the generation time of the site's home page. In this case, 7
seconds.
We looked hard at postgresql.conf, planned the memory usage, sort_memory
and all that. That was a boost. Then we looked at the queries that were
being thrown at the database. Over 200 to build one page! So, a layer
of caching was built into the web server layer. Finally, some frequently
occurring combinations of queries were pushed down into stored procs.
We got the page gen time down to 1.5 seconds AND the server being stable
under extreme stress. So, a fair win.

Thanks to cms for several clues.

So, without understanding your application and were it's taking the
time,
you can't begin to estimate hardware usage.



pgsql-performance by date:

Previous
From: Ron
Date:
Subject: Re: extremly low memory usage
Next
From: Stéphane COEZ
Date:
Subject: Re: Performance pb vs SQLServer.