Archives of the TeradataForum

Message Posted: Fri, 06 Jun 2003 @ 23:42:04 GMT

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

I'd like to make a correction to one of the statements I made yesterday. My apologies if I misled anybody.

David's initial expectation was correct. Transient journaling is for the most part suspended when inserting into an empty partition of a PPI table, assuming PIs of source and target are the same. Transient journaling of individual rows does not take place, but a special transient journal entry is made for the partition as a whole. The insert into empty partitions deletes the partition on a rollback. This applies only if the insert is in its entirety targeted to an empty partition, and does not span partially filled partitions.

There might be some of the other empty table optimizations that come into play in the 8 second example. Empty tables can get optimizations beyond just the suppression of transient journaling - like avoiding spooling and sorting. Further, transient journaling contributes a fairly modest overhead, nowhere near the 4+ times difference seen in David's examples. So transient journalling, or its absence, may not explain the timing discrepancies. The presence of other structures, such as fallback or NUSIs, might.

Thanks, --Carrie Ballinger

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