Archives of the TeradataForum

Message Posted: Thu, 21 Apr 2005 @ 13:24:23 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

<-- Anonymously Posted: Thursday, April 21, 2005 09:23 -->

This is a purposeful restriction and NOT an arbitrary omission.

Duplicate index check for non-PPI tables (or PPI tables that include the partitioning column in the PI) and even duplicate row checking for SET tables all operate under the assumption that a relatively small number of data blocks, all relatively "nearby" (e.g. same virtual cylinder) will have to be accessed for the check.

Of course, it's possible for poor PI choice (highly non-unique and/or badly skewed) to invalidate that assumption and trigger performance problems. But checking for duplicates across partitions would always require accessing at least one block per partition. Using USI instead makes the index duplicate check "local" to a single block again.

	<Prev	Next>		<<First	<Prev	Next>	Last>>