Re: Principle of Orthogonal Design

From: Jan Hidders <>
Date: Fri, 18 Jan 2008 04:51:05 -0800 (PST)
Message-ID: <>

On 18 jan, 02:57, JOG <> wrote:
> On Jan 17, 5:08 pm, Jan Hidders <> wrote:
> > On 17 jan, 15:14, "Brian Selzer" <> wrote:
> > > This calls into question the idea that it should
> > > be possible to determine which relation an inserted tuple is destined for.
> > Apologies for being lazy, but could someone explain to me in a
> > nutshell why this should be possible at all? At first sight this looks
> > like complete nonsense to me.
> Well, as far as I gather, being able to determine a unique predicate
> for any proposition being inserted into the database is desirable in
> order to allow view updates to be more easily be translated to changes
> in underlying base relations.
> I cannot claim to fully understand the reasoning behind this, but view
> updating hence appears to have been POOD's underlying motivation.
> There is a relatively old draft paper by Date and McGoveran at living
> at might be more
> illuminating.

It mentions the view update problem but doesn't really explain how it is connected.

Anyway, the stronger POOD that requires that headers are distinct sounds like nonsense to me. Why would R(A, B) be a worse design than R(R_A, R_B)?

The weaker POOD looks more interesting to me. I even found a published paper about it:

What's interesting is that it seems to address the problem of normalization at the schema level, so not just restricted to one relation, as is usual in normalization theory. You might even formulate a new normal form at that level:

  A schema is said to be in HNF if all database dependencies follow logically from the dependencies at relation level (so domain, tuple and relation constraints) and the inclusion dependencies.

So basically it says that inclusion dependencies are the only needed dependencies at database level. Note the similarity with 5NF (all join dependencies logically follow from the key dependencies) which says that the key constraints are the only needed constraints at relation level. That similarity is not a coincidence. In both cases we are eliminating redundancy. To get to a real normalization theory you would have to identify more concretely the type of database constraints / dependencies we are talking about.

What POOD has to do with HNF is left to the reader as an exercise. ;-)

  • Jan Hidders
Received on Fri Jan 18 2008 - 13:51:05 CET

Original text of this message