Re: A voodoo masterpiece

Howard J. Rogers wrote:
> Holger Baer wrote:

<snip>
> > I tried to reread Mike's comment. However, the page seems no longer
to
> > exist - or maybe it will reexist on the 18th? ;-)
> >
>
> Oh, that's another Voodoo trait: sanitise the record when the record
proves
> too embarrassing. Fortunately, I think Jonathan has saved the
original.

Whilst I know where you are coming from, I also know about putting stuff up before I am ready to publicise it, then finding someone had picked up on a half completed bit of writing (don't you love search engines). Given that the article was dated when it was I think it would be wise to reserve judgement until the article actually appears (assuming that it does).

> > I agree that Mike's statement is valid,
>
> Of course it is. My scorn was reserved for Don's 'proof is beyond
stupid',
> not Mike's 'tests break down on big systems'.
>
> Like you, however, I have concerns that "tests break down" should
never mean
> "so don't bother doing any tests in the first place". I don't *think*
Mike
> has crossed that line yet, but Don certainly appears to have done.

I thought that I'd posted something along these lines yesterday so apologies if this is a repeat.

One thing that seems to be missed in all this is that tests breaking down is , generally a good thing. There will be a specific reason for the test breaking down, and finding that reason will likely tell you something useful about incorrect assumptions that you have made, or factors that you have ignored.

The other suggestion that I'd make is that I at least, but I believe others, tend to post demonstrations rather than 'proofs', for example I have posted a script to demonstrate uniform index height. You cannot use this script per-se to 'prove' that a b*-tree index is *always* balanced - one counter-example would be enough. Obviously as the number of times you run the script with different variables rises and you do not find an unbalanced index the likelihood of the assertion being false diminishes. By its nature scientific 'proof' rarely exists - the scientific method generally will only disprove theories (though there can be exceptions if you can express the theory mathematically). However as the number of times predictions are confirmed by different tests rises so too should your confidence in the theory.

Niall Litchfield
Oracle DBA
http://www.niall.litchfield.dial.pipex.com Received on Wed Nov 10 2004 - 03:16:23 CST