Re: Dreaming About Redesigning SQL

In article <[EMAIL PROTECTED]>, Patrick K. O'Brien
<[EMAIL PROTECTED]> writes
>Lauri Pietarinen <[EMAIL PROTECTED]> writes:
>
>> Now we really come to the core issue because you will end up going
>> thru all OrderDetal rows even when only a subset would be necessary.
>> Performance would be unacceptable. You should start from either
>> Customer or Product table, depending on how many customers there are
>> from London or products in group 'Screws' there are.  And this
>> information is available to the optimiser of the DBMS.  The user
>> merely needs to run some kind of statistics updating from time to
>> time (or, in the case of SQLServer, not even that).  So on Monday,
>> the acces path could be different from that on Thuesday, or, even
>> depending on the input given by the user of the application.  And
>> all these different access paths originating from one single
>> SQL-query.
>
>As I've said before, I haven't focused on optimization yet.  But I
>remain convinced that an ODBMS with the proper object model can be
>optimized as you describe.  Does that make me an optimist?  ;-)

Why is everybody messing about optimisation? Why doesn't everyone use a
database that DOESN'T NEED to be optimised :-)
>
>> >Clearly that approach lacks the generality of a good, ad-hoc,
>> >declarative query language.  And it doesn't take advantage of the
>> >indexes.  That wasn't the original argument I was trying to make.
>> >I also haven't spent much time on performance optimization.  But I
>> >will get there over time.
>> >
>> It has taken 30 years of (still ongoing) research for the products
>> to get to where they are now.  Do you really think you can get there
>> "over time"?
>
>Yes, I do.  Why not?  Why can't the results of that research be
>applied to an ODBMS?
>
I couldn't agree more.

Just look at Linus' comments when he first started linux. All the
research was going into micro-kernels. All he did was have the (not
particularly bright) insight that all the optimisation that was being
done to microkernels applied just as much to macrokernels. And
macrokernels have the inherent advantage that they don't context-switch
- a very expensive thing to do. So a macro-kernel, by its *very*
*nature* will always outperform a micro-kernel, if properly optimised.
(Just like a good big camera will always beat a good small one - the
laws of Physics say that is inevitable...)

What's the betting - all this research being thrown at relational is the
database equivalent of micro-kernel research :-)

Cheers,
Wol
-- 
Anthony W. Youngman - wol at thewolery dot demon dot co dot uk
Witches are curious by definition and inquisitive by nature. She moved in. "Let 
me through. I'm a nosey person.", she said, employing both elbows.
Maskerade : (c) 1995 Terry Pratchett