home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!keele!csa09
- From: csa09@seq1.keele.ac.uk (Paul Singleton)
- Newsgroups: comp.databases.theory
- Subject: Re: Normalization and Relations
- Message-ID: <4052@keele.keele.ac.uk>
- Date: 11 Sep 92 13:38:18 GMT
- References: <gwtn05l.sjs@netcom.com>
- Organization: University of Keele, England
- Lines: 49
-
- From article <gwtn05l.sjs@netcom.com>, by sjs@netcom.com (Stephen Schow):
- > There is a methodology called "Extended Relational Analysis" that goes into
- > how to design a relational model methodically.
- >
- > It basically talks about identifying the entities of your model. You can
- > generally use nouns to refer to these entities. Each of these entities
- > will be represented by a table.
-
- Shouldn't this be: "Each of these entities will be represented by a unique
- identifier"? Simple attributes (e.g. 'name') of these may indeed be stored
- in a table. It seems to me that traditional relational analysis likes to
- confuse an entity with a (hopefully) distinguishing attribute of that entity.
-
- I think "functional data modelling" is more careful about this, but I haven't
- studied it.
-
- > ...
- > What you have to be careful of is that you could practically call everything
- > a noun and get totally out of control trying to normalize your data. As has
- > been said earlier by someone in the group, if a particular entity is not of
- > unique interest by itself, or if it does not stand on its own as an entity,
- > then it probably shouldn't be one.
-
- Sorry, why would you "get out of control"? You might get close to the truth.
-
- > There are many exceptions. If you are talking about gobs and gobs of redundant
- > data, then it may be worth it. If you are talking about a one to many
- > relation, then it is neccessary. There are many variables. But I think it
- > is key to try to find a balance between absolute normalization and a more
- > conservative approach.
-
- I think it is misguided and hopeless to "try to find a balance ..." without
- *first* analysing the application data thoroughly and uncompromisingly. I
- think there should be two data models:
-
- 1) an exhaustive and fastidious analysis of the application, e.g. with
- attributes such as 'colour' being treated as entities unless we can
- be absolutely sure they are OK as attributes;
-
- 2) a pragmatic representation schema, deliberately denormalised for
- efficiency and compactness (if necessary), and with all compromises
- *documented* and *rationalised*.
-
- But I've never done this for a living ...
- ----
- __ __ Paul Singleton (Mr) JANET: paul@uk.ac.keele.cs
- |__) (__ Computer Science Dept. other: paul@cs.keele.ac.uk
- | . __). Keele University, Newcastle, tel: +44 (0)782 621111 x7355
- Staffs ST5 5BG, ENGLAND fax: +44 (0)782 713082
-