triple|scape

Bob DuCharme put out a call for questions to be addressed during the Linked Data Workshop at LinkedDataPlanet. The best question gets the author into the conference for free! So, I submitted a pretty open-ended question with very low hopes of winning back my $900-ish registration fee. Probably won’t win (though I’m the only submitter so far). For fun, I thought I’d post my question here and see what you think:

This is a fairly broad question, but: When it comes to exposing enterprise data from proprietary and COTS systems in a linkable way, two obvious choices come to mind: Provide something like a SPARQL front-end on front of each application, or aggregating all of that into a queriable “semantic data warehouse”. What are the benefits and pitfalls of those approaches, and are there approaches that lie somewhere between that have proven to be successful?

Thoughts? Comments?