I assume everyone understands the need for a unique clustered index. on almost every table.  Without a clustered index the table is just heap.  You have no easy way to cleanup fragementation without a clustered index. If you do not make the clustered index unique SQL Server creates a uniqueifier for you to make the index unique.  The old school of thought was to cluster for distribution to ease hot spotting on inserts.  Since row level locking has been implemented the need to distribute inserts has become secondary to minimizing index and extent fragmentation.  If  you use an identity column as your unique clustered index you can greatly reduce the amount of fragementation because all of your inserts will occur at the end of the table.  You need to experiment to find a good fill factor for you particular situatiuon to minimize page splits from updates.  In the sessions on indexing I attended it was suggested that you fill varchar fields to some average default length on insert to help reserve some space and miniminize page splits.  I know I am going to try implementing these suggestions in the next systems that I design and implement.  Many of these suggestions raise the old issues of natural or surrogate key for primary keys.  After the session I am now more convinced that I need to consider how sql server works more and how can I usew that to build high performance systems.