Prompt:
Big central warehouse, or local datamarts? That was once the question.
The cultural challenges of data management seep between the lines of this article. No catfight is fiercer than two modelers defending their models.
Data people get into needless religious wars because in action, all data is local, but until action, it is all central, "bloody, perjured, murderous, full of blame." Or to paraphrase the old philosophy teaser, if the data is stored but there's no one there to use it, is it data?
When a canonical source can be farmed out to local users who then cycle back to the source with feedback, and if the dw has a quality adjustment procedure for handling this feedback without religious wars, you have a coherent data system that can evolve with business climates.
Certain radical db modelers (read about one at fluidinfo.com) despise ontologies for this reason: they revere the chaotic coherence provided by local users' management of the data, by analogy with biological fitness landscapes: the fittest data shall survive.
But the beauty of the DW marriage with the data mart, if united in marriage by a data quality framework, is that the survival doesn't have to be brutal. Where the core model is stable (think of the 90%+ of DNA among vertebrates in nature, or rule of law among societies), and it is allowed by a "data constitution" to evolve (think of American democracy, however ugly it may look at present), these two approaches can allow local users to develop and maintain their data, and central modelers to learn from the masses.
The big win here, though, is not for the enterprise--the corporate entity--, which may be on the ropes as the global economy changes. It is for the agile enterprises that arise and dissipate to meet market needs. One can imagine persistent data stores that exist independently as core data libraries "rentable" to corporate entities that appear and disappear, developing their own marts as clones of the central one.