Re: Decline of Science: Computer Science and Databases

From: Carl Rosenberger <carl_at_db4o.com>
Date: Fri, 29 Nov 2002 17:33:32 +0100
Message-ID: <as84ci$cob$02$1_at_news.t-online.com>


Eric Junkermann wrote:
> > Your generalisation certainly is false for usecases where
> > - a network model is used to hold data
> > - compatibility to SQL is not an issue
> > - only a hierarchical access path is necessary
> > - time to market is essential
> > - the class model may be constantly refactored and reducing
> > maintenance work to do so is a must
> > - navigation solves all issues perfectly and queries aren't needed
> >
>
> So you are constantly refactoring, and you want to reduce maintenance
> work. But presumably your network model and your hierarchical access
> paths are tailored to your class model, and will also have to be
> refactored

Certainly.

> - how much work is this, how much data conversion do you have to do?

Ideally this is zero work.

The hierarchical access path *IS* the class model. The class model is the only model that is needed. It defines the data structure.

The database vendor supplies a plugin for the integrated development environment that understands refactorings. It creates a log of the refactorings applied. This log can be deployed to running applications.

> How much data manipulation is needed to upgrade a live
> data store to the next version of the application (and how thoroughly
> can you test it)?

The update works as above.

Testing is a different beast. How thoroughly can you test that an application will survive your SQL UPDATE statements?

> And, most importantly, how do you deal with a new requirement that
> does not fit neatly into your current access paths? Or to put it
> another way, how do you know that your rigidly defined navigation will
> solve all _future_ issues?

Class models are always alive.

Access paths can be extended.

Furthermore there is no reason that object databases should not support relational queries. They will just not be as fast as direct navigational access "pointers".

Furthermore there is no reason that object databases should not be able to supply updateable object views. (We do.) It's easier to map objects to objects than it is to map objects to tables.

Kind regards,
Carl

--
Carl Rosenberger
db4o - database for objects - http://www.db4o.com
Received on Fri Nov 29 2002 - 17:33:32 CET

Original text of this message