Re: A Normalization Question

From: VHarris001 <vharris001_at_aol.com>
Date: 09 Jul 2004 13:19:05 GMT
Message-ID: <20040709091905.29796.00001160_at_mb-m18.aol.com>


Tony wrote:

>
>Perhaps you might like to build an XDb1/TDM physical implementation of
>an RDBMS like that. It would have a big index file full of values
>like this:
>
>Index# Value
>1 "brown"
>2 3.1415926535
>3 DATE '2004-07-07'
>4 "blue"
>5 DATE '2004-07-01'
>6 "Mary"
>...
>
>When the user inserts a row into a table using SQL like this:
>
>INSERT INTO person (name, eye_color, date_of_birth) VALUES
>('Mary','green',DATE '1970-01-01');
>
>... your RDBMS can look up the values 'Mary', 'green' and DATE
>'1970-01-01' in your humungous VALUES list and either return the
>index# if it exists, or create a new entry and return its index# if
>not. Then it can store the record using only your index# pointers
>like (6,42,10232).
>
>For a SELECT, a similar process will be invoked.
>
>I suggest you go for it: clearly you think this will be a better
>RDBMS, and the SQL interface will make it marketable.

Well, the RDBMS either has to store it and (likely) index it, or look it up in the index and point to it. How much difference could that possibly make?

Besides, there's another advantage of the single table of things' in that only populated fields will be in the table. The things table wouldn't be as large as the pointers table. That, I presume would be enormous.

V Harris Received on Fri Jul 09 2004 - 15:19:05 CEST

Original text of this message