Archives of the TeradataForum

Message Posted: Mon, 07 Apr 2014 @ 20:56:00 GMT

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

When you do single-row inserts, the PE will fetch and increment the IdCol value as specified in the definition.

But this would be a huge bottleneck for all-AMP operations (including load utilities and INSERT/SELECT), so a different approach is used. Each AMP in turn fetches a starting value and increments it by the IdColBatchSize value from dbscontrol (default 100000), effectively "reserving" a batch (i.e. range) of identity values for that AMP to use. Values are then locally assigned to rows until the batch is used up, at which point the AMP would reserve another batch of values. Values reserved but not used in the load are effectively "lost" and will not be assigned (unless you allow CYCLE, in which case you lose uniqueness).

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