Archives of the TeradataForum
Message Posted: Thu, 05 Jun 2003 @ 23:02:20 GMT
I don't have all your answers, but I think I can help with one or two of your questions.
The optimized insert/select is only available when the entire PPI table is empty. If one partition contains rows, inserting into a second, empty partition will still incur the overhead of transient journaling. That is what you are seeing with your second of three inserts.
However, transient journaling is suspended when deleting from a partition. One transient journal entry is written to delete the entire partition when the transaction commits.
None of the plans I have seen for insert/select from NPPI to PPI table when the primary index is identical have shown row redistribution, as your plan below does (neither do such plans make references to the number of partitions involved). I've only ever seen row redistribution when the PPI table has a different PI from the source table. The partitioning differences between the source and target tables do not matter in this regard because they are AMP-local differences and do not require moving rows to different AMPs to resolve. Unfortunately, without the two table layouts and the actual SQL used, it's difficult to say why you might have gotten redistribution in your plan.
Thanks, --Carrie Ballinger
|Copyright 2016 - All Rights Reserved|
|Last Modified: 27 Dec 2016|