On 10/4/07, John Stanton <[EMAIL PROTECTED]> wrote: > A B-Tree index holds keys in sorted sequence. They are in random > sequence in the database. That requires holding the keys in the B-Tree > nodes.
Actually, it doesn't strictly require that; it could store references to the keys. An obvious tradeoff is I/O; an index walk is less useful if you have to do random seeks to the locations of row data just to get the keys to walk the tree in the first place. IOW in simplistic terms, an index walk suddenly doubles in disk I/O. The information on SQL Server would be interesting, as I know it stores sort keys under some conditions, which is effectively duplicate data. ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
