Very nice overview. Gail, one thing that I was wondering if you could comment on... Towards the end there was a discussion of the best type of clustering key to use. I totally understand you're providing some very good general rules, but one of the attributes was that it is unique. Often times a very important use of a clustered index is to organize data in the detail table of a summary-detail table pair, for example, orders and order items. The most common query against the detail table will be to find those rows that correspond to a given summary row (ethier when looking up a specific order, or aggregating detail rows for all or a group of summary rows). In this case it is terribly important to use the one and only clustered index in the detail table to be on the foreign key linking back to the summary table, I think? I'm pretty sure that's generally the right way to do things so that range queries on the detail table (again, assuming they're the most common queries on the detail table).

Anyways, I just thought that since summary-detail table relationships are very common that it's probably worth calling out this exception to the uniqueness guideline specifically. What do you think?