I'm current working on a 2-year data architecting project of a size, scope and visibility that could produce severe bowel control problems if we don't get it right. Aside from the enormity of the data modeling task - even with tools - the old religious war of natural key vs. surrogate key is again raising its ugly head. Some spot research on the latest thinking revealed a nicely-written commentary on the battle and merits of both.
I respect Walker's approach to the topic as rejects the usual dogmatic standpoints of the extremists, and rational explanation of his views. While I'm not 100% convinced on broad use of surrogate keys I appreciate the article's points of data abstraction and performance considerations. However, as with a lot of articles on the topic it neglects the downstream issues with identity columns (table replication/loading, archiving, etc.) but those are a minor concern. Overall I liked the arguments and will rely on Walker's points when I get to the physical model.