Archives of the TeradataForum

Message Posted: Tue, 30 Nov 2004 @ 14:35:15 GMT

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

Hello Phil,

Thank you very much for sharing this information. The idea of having subsets of the data in "cyclical" tables sounds interesting. Al though our users are free to select any time-period for analysis - we use MicroStrategy as end-user tool.

We are on R5.0.3 and uses PPI in some cases. We will soon (hopefully) upgrade to R5.1.1, which has dynamic partition elimination on product joins. This should encourage a PPI-design with PI = (basket_id,date), partition on date. Maybe that could eliminate the need for subset-tables of the kind you describe?

Any way, if the design is normalised, the PI of header and detail of course must be the same. So that leaves (basket_id), (basket_id,date) or perhaps (store_id,date) as candidates. The latter two only being relevant for a PPI- design. The interesting thing about (store_id,date) is that it would also enable a star-join, and with PPI, R5.1.1 has a very efficient rowkey based merge join. But it would probably need a Single table Join Index on the detail table with PI = (basket_id) to support affinity analysis.

Best Regards,

Ole Dunweber
Coop Nordic

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