I tend to focus on High Performance databases where performance is critical. Aaron Bertrand suggests skipping a PK even while having a cluster in his ActivityLog example. From a performance aspect not having a PK strikes this writer as foolish. I tend to like the "ID INT IDENTITY(1,1) PRIMARY KEY," style since it takes half the index space as making the Datetime a PK. It tends to be a trade off, I like to use more storage space for faster performance, some people will trade performance for a little cost savings.

Best wishes,

Barry O’Connell