• From: "Peter Hunsberger" <peter.hunsberger@g...>
• To: "Len Bullard" <len.bullard@u...>
• Date: Thu, 15 Nov 2007 16:05:02 -0600

On Nov 15, 2007 2:31 PM, Len Bullard <len.bullard@u...> wrote:
> Dave also sent me that.  To do otherwise would have been to give DTDs a
> longer lifespan.  It confirms comments made on this list along the way about
> the necessity of creating minimal standards and really, minimal systems.
>
> KISSGodzRule:  Any feature acquires a use case somewhere and becomes almost
> impossible to remove post hoc.
>
> As I sit here planning a 'features complete' release of a CAP-conforming
> health alerting system, I marvel at the truth of that.
>
> Here's something for you to ponder:  what percentage of a database entity
> design (and consequently, supporting forms, business classes etc) should
> support domain-lookups (eg, metadata) vs operational data (ie, business
> data)?  It's not a trick question but the answer affects product planning
> and standards planning.
>

I had sent this reply to Len off list, but it's sort of an interesting
debate so here it is again:

If you're collecting research data then, off hand, I'd say about 90%+:
 we're using a strongly typed version of a EAV model, only 5 tables
out of the 40 or so involved in the model actually collect the actual
data.

I had figured Len would come in high but not that high, instead he
agreed.  I actually think we can drop that ratio by recursively using
the data tables to store the metadata; sort of an uber schema. There
seems to be a belief in some circles that such a design result in a
lot  of extra joins, but in my experience if it's done right that's
not particularly true...

-- 
Peter Hunsberger

• References:
  • RE: 2 questions about Relax NG: ID/IDREF datatype and whitespace facet
    • From: Len Bullard <len.bullard@u...>