Re: Data Model - Please Comment on my Approach
From: Karsten Farrell <kfarrell_at_belgariad.com>
Date: Fri, 23 May 2003 23:18:56 GMT
Message-ID: <MPG.19384c3ab22de54c989799_at_news.la.sbcglobal.net>
Date: Fri, 23 May 2003 23:18:56 GMT
Message-ID: <MPG.19384c3ab22de54c989799_at_news.la.sbcglobal.net>
71062.1056_at_compuserve.com said...
> The rest of your design is a meta-data
> model, with column descriptions. You will wind up with monster joins.
> Sorry, but this is not a good idea at all. The original model is
> correct.
>
Well, I have two comments:
- There is nothing wrong with joins. They are a part of life in the
RDBMS world. You normally eliminate many of them when you build a Data
Warehouse, but ONLY then. One of the most common type of joins is done
to "convert" (display) a code to its description (they're called lookup
tables, for obvious reasons).
- The original model is not even first normal form (1NF), but is reminiscent of the flat-file databases of the 1970s. Normally you try to get your model to at least 3NF. To get to 1NF, you eliminate all the column "arrays" - even though they're not called that in a table. The reason the original model is flawed becomes obvious when someone decides to breakdown age brackets into smaller (or larger) groups than the current 5-year groups ... or they want to capture other income levels ... or they want to toss in another house type. If you have stored nothing but totals, you'll have no way to determine which of the tens or thousands now fall into the split/joined categories.
--
/Karsten
DBA > retired > DBA