How to Think Like the SQL Server Engine, Part 2: Nonclustered Indexes