The fact is, all short comings of either approach are manageable. Just use views and surrogate keys. This argument happens everytime this topic comes up. The real issue is which approach has less problems. Views don't take considerable time to load, unless you're trying to do something stupid like pull entire multimillion record tables to the workstation. The old fox browse never did that. It had a means of extracting just enough data to display until you "requested" more by scrolling down.

In your example of sorting by the description field, how many records and fields are in the tables? A view can be indexed on the description by the client machine. That precludes a need to maintain another index that slows performance when these indexes are to be updated.

There are basically two approaches. 1) the surrogate meaningless keys 2) the meaningfull keys. Each approach has problems If you built a line item with this meaningful key...

InvoiceNumber+LineNumber

If the invoice number changes, you are not just doing a simple cascade replacement, but a replacement of part of the key.

Walter was suggesting that we don't think about this stuff. That's because supporting both ways of accomplishing a task is inefficient. Rehashing this topic over and over is also inefficient. I was very frustrated by the grid's lack of Rushmore Optimization support, but now I use views and I only pull the data people need and you know what? I don't care anymore. Performance is good and users are happy.

It seems more sensible to master one approach.