Performance - Mailing list pgsql-hackers

From Turbo Fredriksson
Subject Performance
Date
Msg-id 87sn95tba4.fsf@papadoc.bayour.com
Whole thread Raw
List pgsql-hackers
I was reassigned (temporarily thank you very much! :) to the test
department at work to do some performance tests on our PostgreSQL
(v 7.1.3) database.

I've seen quite a number of test/benchmark programs on the 'Net, but
none that fits the purpose...

We all know that 'there is no problem with speed on a PostgreSQL database',
but still the other developers at work is pointing fingers at PostgreSQL...

Now, _I_ say that it have something to do with the _USE_ of PostgreSQL, not
the actual database per see.

The question is, how do I test _OUR_ database? That is, with our database
schema, function etc in a 'compliant' way? Some people keep talking about
measuring the CPU load, but I keep telling them that that's moot information!
The values is basicly useless (for a number of reasons).


Our setup is based on three components, called 'The VXIS', 'The VSUS' and
the database. It have to do with SMS (I can't tell you much more than that).

Anyway, the SMS is arriving from the cell phone to the VSUS which stores the
SMS in the database. The VXIS is retrieving the SMS and delivers it to the
SMS Central.

There can be a number of VXIS' and a number of VSUS', depending on requested
performance. Now, we have tested to send SMS'es between two VXIS' and we
are getting around 7000 SMS/sec. That's by _NOT_ going through the database.
But when we add the database, it drops down to 100 SMS/sec one way, and 40
SMS/sec the other way!!

What we now have to do is figuring out _EXACTLY_ where the problem lies. Is
it because of stupid/slow queries? PostgreSQL setup/compiled wrong? Etc, etc.
I keep telling everyone that it CAN'T be a problem in PostgreSQL, but have to
be somewhere else. But my bosses don't care, they need number (darn PHB's ! :)


But I need a 'correct number' to compare the different tests - Did the change
do any improvement?

How do I arrive at this 'correct number'?

As you can hear, I'm no competent tester, I'm a developer. I usually don't have
to test my product that much. I always leave that to the user :) Could anyone 
share there wisdom, and give me some pointers?

-- Turbo     __ _     Debian GNU     Unix _IS_ user friendly - it's just ^^^^^    / /(_)_ __  _   ___  __  selective
aboutwho its friends are         / / | | '_ \| | | \ \/ /    Debian Certified Linux Developer   _ /// / /__| | | | |
|_||>  <  Turbo Fredriksson   turbo@bayour.com \\\/  \____/_|_| |_|\__,_/_/\_\         Gothenburg/Sweden
 

genetic smuggle NSA FSF Peking cryptographic Cuba fissionable Rule
Psix Qaddafi CIA Serbian World Trade Center Clinton colonel
[See http://www.aclu.org/echelonwatch/index.html for more about this]


pgsql-hackers by date:

Previous
From: Jean-Michel POURE
Date:
Subject: CREATE OR REPLACE VIEW / TRIGGER
Next
From: "Marc G. Fournier"
Date:
Subject: Re: RC1 date?