Re: 4 the FAQ: Are Commercial DBMS Truly Relational?

From: Andrew McDonagh <news_at_andrewcdonagh.f2s.com>
Date: Mon, 11 Oct 2004 23:12:30 +0100
Message-ID: <ckf0gq$vkp$1_at_news.freedom2surf.net>


Christopher Browne wrote:
> Clinging to sanity, Andrew McDonagh <news_at_andrewcdonagh.f2s.com> mumbled into her beard:
>

>>Laconic2 wrote:
>>
>>>"Paul" <paul_at_test.com> wrote in message
>>>news:41690aad$0$59441$ed2619ec_at_ptn-nntp-reader03.plus.net...
>>
>>Snipped
>>
>>
>>>There is a widespread belief in this field that wrong answers with
>>>good performance are closer to the goal line than correct answers
>>>with poor performance.  I will never come around to that point of
>>>view.  I almost always want to get it right, first, then work on
>>>getting it right, and fast.
>>
>>I agree, its usually a case of 'premature optimisation'... i.e. they
>>use these 'tricks' for optimisation thinking that will create
>>performance benefits, but usually the real bottle necks are else
>>where within the system. Therefore applying these tricks only serve
>>to complicate the matter.

>
>
> The notion of "wrong answers with good performance" can also fit with
> the notion of using approximations.
>
> There are cases where modelling the exact correct result is costly,
> and that an approximation will suffice at lower cost.
>
> There are numerous sorts of costs:
>
> - CPU usage
> - I/O usage
> - The amount of time and skill required to generate the report
>
> An approximate answer that takes an hour to find that is "close
> enough" that allows timely decision making is better than one that is
> exact but that took too long to find so that the information arrived
> too late for use.

Horses_for_courses

If there's a requirement that says approximations are fine, then yes you are correct.

Or if there's a requirement that correct (i.e. no approximations) is required but must 'run like the wind', then the design can and probably will be changed to accomplish this.

But, if neither requirement is made, then the design should always over rule speed.

And from my own experience, accomplishing the correct answer quickly, can usually be achieved whilst still maintaining a good design. Usually the design is different to the original one, but it is equally 'good'.

Andrew Received on Tue Oct 12 2004 - 00:12:30 CEST

Original text of this message