Archives of the TeradataForum
Message Posted: Wed, 29 Mar 2000 @ 18:19:29 GMT
Just briefly, consider having a NUPI as region, store & date, i.e remove tranid. Also, try to 'encourage' your users to navigate the tables via the reference data, in particular try to make them qualify on all three NUPI columns. The reason for this is that it is highly unlikely that a user is going to ask a question like "Sales for region 2, store 1, transaction date = '20000301' and tranid = ???", in which case the PI will not be used. Also, by navigating via the reference data the optimiser is more likely to use a star join.
If you treat the detail tables in the same manner, then any 'basket' analysis that is performed, even though tranid etc will be required for the join, the optimiser may still decide to use the NUPI.
Hope this helps
|Copyright 2016 - All Rights Reserved|
|Last Modified: 27 Dec 2016|