Re: A kind of... benchmark?

From: Holger Baer <holger.baer_at_science-computing.de>
Date: Thu, 07 Aug 2003 08:42:20 +0200
Message-ID: <3F31F4CC.3070605@science-computing.de>

Billy Verreynne wrote:
> Rick Denoire <100.17706_at_germanynet.de> wrote :
>
> <snipped>
>

>>Please be aware that I am not trying to know anything about
>>performance of a specific database (meaning the schema and data
>>contained therein), but about performance of database *environments*,
>>i.e., the DB engine and its parameters, OS I/O system, memory
>>management, CPU performance etc. I am interested in the administrative
>>aspects, while efficience of work with specific relational models
>>would be something related to database development.

>
>
> Not sure if I follow the logic here...
>

<SNIP>

>
> What will such a benchmark prove? That one machine is "better" than
> another? That is pointless. Not only does technology continually
> change, but business requirements are not fixed either.
>
> Benchmarks are useless IMO. They serve no purpose but as a penis
> enlargement for marketing & sales. It is totally meaningless in the
> real world.

<SNIP>

Although I generally agree with your statement, I believe there is actually one thing where benchmarks are usefull: Learning the database.

With the benchmark, you have a environment with complete runtime metrics, something that *real* applications often lack. So in this artificial environment you have not ontly the database metrics you can gather with statspack or set events 10046 and tkprof, but also metrics more from an applications point of view. With the combined information carefully analyzed you can try different strategies for setting up your database (oh, and the benchmark of course), and get rid of some the oracle myths as you go along.

But then, this is not what the OP was after.

My 2c

Holger Received on Thu Aug 07 2003 - 01:42:20 CDT