Archives of the TeradataForum

Message Posted: Fri, 26 May 2006 @ 14:06:42 GMT

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

<-- Anonymously Posted: Friday, May 26, 2006 08:50 -->

The manual offers good guidance but is not a substitute for experience.

Denormalization is best left to Data Architects that can weigh in the benefits of denormalization against the maintenance requirements of a denormalized model and the possibility of change. The goal is not "pure performance" but a balance between "performance", "maintenance", and 'managing change"

In my case, Teradata technology and speed makes it a very easy choice. My DW is strictly 3NF. ETL is a straight forward process and Data Mining takes place there. A set of Dimensional Models implemented using fact tables and 26 Join Indexes provide the keys to servicing the needs of over 4000 daily promptable MSI reports, many of them are YTD reports using This Year Last Year comparisons. The use of Join Indexes makes the Dimensional Model maintenance mostly a non issue because the Join Indexes make it "automagic" (i.e., doing updates, deletes or inserts into the fact tables).

Needles to say, the machine is at idle most of the time giving ample room for data mining queries on the 3NF side.

Anonym

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