Archives of the TeradataForum

Message Posted: Tue, 20 Apr 2005 @ 00:09:08 GMT

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

One reason you would not want to define a PI as a UPI when the table is partitioned, unless all partitioning columns have been included in the PI, has to do with the overhead of enforcing uniqueness. In your example, if (a,b) were defined as a UPI, then for each insert of a new row, each partition on the AMP the row hashed to would have to be examined to see if a duplicate value for (a,b) already existed. With a large number of partitions, that overhead of probing each partition during duplicate checking could become problematic.

If you really do required (a,b) to be the PI and you would also like its uniqueness to be enforced, one option is to create a USI on (a,b). That way uniqueness will be enforced at the time of an insert, but it will not require probing each partition.

Thanks, -Carrie

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