Archives of the TeradataForum
Message Posted: Wed, 18 Oct 2000 @ 12:15:27 GMT
I would suggest that you also look at the your primary index and decide if you really need all 8 columns in the primary index. You mention that you have 50K customers and 6 million rows. If you made the customer ID the primary index, you would end up with an average of 120 rows per PI value (you would also want to determine the max). If you use MULTISET tables, the performance penalty for having a high number of hash collisions goes way down. So, you may want consider making your primary index a NUPI with just customer ID or customer ID, combined with some other column that has high number of possible values.
By having fewer columns in the primary index, the optimizer will more readily use the "star-join" path of product joining the dimensions to come up with the PI values for the fact table. If you are able to reduce the PI to just customer ID, the optimizer wouldn't even need to employ the star-join path, and could quickly access all of the rows for a particular customer.
Hope this helps.
|Copyright 2016 - All Rights Reserved|
|Last Modified: 28 Jun 2020|