Re: Pizza Example
Date: Mon, 19 Apr 2004 20:35:44 +0100
In message <slrnc84bpk.qnf.choess_at_force.stwing.upenn.edu>, Chris Hoess
>>>So statements in a relational database are guaranteed to be true to the
>>>degree that truth can be defined by the constraints expressed in the
>> Substitute "truth" with "consistency". If, by "truth", you mean
>> "corresponds with reality", then (a) you can NOT do it automatically,
>> and (b) logic is completely the wrong tool - after all didn't Aristotle
>> prove that a pound of lead is heavier than a pound of feathers :-) (yes,
>> I know I've quoted the story wrong - it was something about falling
>> faster or something :-)
>Well, it corresponds with reality insofar as the constraints in the database
>represent those of reality, and the divergence between the two may be very
>significant. But if you think logic is the wrong tool, then the issue goes
>far deeper than the relational model versus the "PICK model" versus whatever
>Neo peddles; computers aren't very good at doing things other than logic, so
>if you throw that out the window, we're pretty much back to flat files.
No we're not back to flat files. Logic is USELESS at proving something
WORKS. But without logic we don't know what to TRY. That's what winds me
up about so many relational proponents - they use logic to prove their
ideas are *consistent*, and then just assume that because it's
consistent it must work. As I've said many times, Newtonian Mechanics is
easy to prove with logic. It just doesn't work in practice.
>>>I am neither by training nor practice a logician, so I'm not entirely sure
>>>this description is correct, but I think it's a good starting point for
>>>discussion of the issue.
>> By training I'm a scientist. Logic is a wonderful tool, but it should be
>> tested by experiment ...
>But I think we're trying to test things at the wrong level. By taking
>sufficiently careful measurements and running them through Newtownian
>mechanics and relativity, relativity should yield a result closer to the
>measured values than Newtonian mechanics. I don't see how one would be able
>to carry out a similar comparison with two database models (although I
>suppose such a test does impose stringent requirements for consistency,
>which may not be true of all of the half-baked stuff put forth). Dawn's
>efforts to put this on a scientific footing, while interesting, are mired in
>a swamp of confounding variables; much more like epidemiology, say, than
The problem is, that by failing to carry out those experimental tests we have no grounds for claiming that the relational model even *works*, let alone is any good, for storing data about the real world.
And the problem you have, is that Dawn and Ross, who have experience of both Pick and relational, say that what evidence they have is that Pick is cheaper, faster to develop with, needs less hardware, and generally works better. I don't have deep personal experience of both, but EVERY comparison I've EVER come across, says that the relational experts have difficulty tuning their systems to match the Pick system. I've already mentioned the consultants who were proud their dual-800 could just beat the old P90 - another little war story is the company who's PHBs told the Pick people "your system is legacy, we need to upgrade, we're going to rewrite in relational". 18 months and $1.5M later, the new system was still not ready to go live and the Pick system was collapsing under the load.
The Pickies went to management and said "we need to upgrade NOW! Why don't you give us the chance you've denied us so far to prove that Pick can do it?" Three *weeks* later, the upgraded Pick system went live.
Why don't I ever hear any stories that say the new relational system was better than the Pick system it replaced. All the evidence I've EVER met has been totally one-sided. Surely that says *some*thing?
-- Anthony W. Youngman - wol at thewolery dot demon dot co dot uk HEX wondered how much he should tell the Wizards. He felt it would not be a good idea to burden them with too much input. Hex always thought of his reports as Lies-to-People. The Science of Discworld : (c) Terry Pratchett 1999Received on Mon Apr 19 2004 - 21:35:44 CEST