Archives of the TeradataForum

Message Posted: Thu, 03 Jul 2003 @ 12:43:10 GMT

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

Hi

Without knowing how large the pdata.shopping_trans table is, I can only make assumptions:

The pmart.fact_sales is a set table with a non unique primary index (NUPI). When you insert a row to that table, the primary index of the incoming row is compared to any rows in the table with the same rowhash. So, if there are 100 rows there with that NUPI value, there will be 100 duplicate row checks. When the next row with that NUPI value arrives, 101 checks will occur.

This can result in the rate of row insertion being reasonable at the beginning and the rate slowing down as the number of duplicate checks increases.

A Multiset table will allow you to overcome this problem BUT you need to be sure that you do not load the data twice and end up with data integrity problems.

Secondly, you may need to see if the table is skewed. This can happen when you have a very common value for a primary index - such as NULL. One AMP may be doing most of the work and you do not realise the value of the parallelism of the Teradata environment. If this is the case, you may need to consider the choice of primary index.

I hope that this helps.

Frank

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