Re: Flamewar object databases vs. relational databases

> > As it stands, I see no end in sight with Carl point-blank refusing to
> > read a relational textbook and an increasingly annoyed Bob descending into
> > increasingly terse profanity.
>
> Actually, both sides have completely missed the point. No relational
> database is going to beat a properly configured OODBMS which has been setup
> for a specific applicaiton with specific access patterns.

You will have to pardon me for thinking otherwise. No OODBMS is going to beat a properly configured relational DBMS which has been setup for a specific application with specific access patterns. A relational database can use the exact same physical storage scheme as the OODBMS.

> Think about it
> ... that schema is typically designed for an app that knows precisely how it
> will access the data

You mean: "... that *physical* schema is typically...."

> RDBMSs by their very nature separate the app logic and data
> storage model, so you're going to get impedance mismatch no matter what you
> try.

This does not cause impedance mismatch. It causes physical independence making the performance characteristics independent of the logical presentation.

> So why don't OODBMSs rule the world? Simple: They aren't appropriate for
> enterprise data stores.

For the same reason that network model databases of all kinds don't rule the world.

> Essentially, an OODBMS data schema tends to work
> well for single applications, or applications that are closely-related.

Even for single applications a network model OODBMS does not work well. I know. I've worked on single applications that used network model databases.

> All this talk about gathering specific record IDs and passing them to and
> fro in some vain attempt to match OODBMS access speed is ridiculous.

I don't recall seeing that talk. I'll have to look again.

> The
> power of an RDBMS is the underlying mathematical model that allows it to be
> an enterprise data store.

I would say that the power extends beyond the mathematical model. In the end, the power derives from application of sound database management principles.

> Given a mission-critical embedded system that needed speed, I'd pick an
> OODBMS in a heartbeat.

If you had a decent relational DBMS available, you wouldn't even consider a network model OODBMS. Unfortunately, until people realize this, insufficient demand will exist for any vendor to actually deliver.

> (Also, while Carl is reading about relational algebra and tuple calculus,
> Bob needs to read the series of responses to the Third Manifesto ...)

I haven't seen those. Where are they?

Regards,
Bob Received on Tue Jun 05 2001 - 00:29:43 CEST