Archives of the TeradataForum
Message Posted: Mon, 05 Nov 2001 @ 10:14:53 GMT
David Roth makes some very good points about 3NF. These points form the basis of my belief that 3NF is usually a good place to start (note I didn't use the word 'best'). As stated in the current edition of Teradata Magazine, the #1 objective of an RDBMS is to separate the data structure (schema) from the application(s) that access the data i.e. aim to model the real world not to pander to shortcomings of the tools/users/developers that will access the data.
I have recently experienced dimensional overlays for MS reporting, both virtual and physical. The tool, seemed to really struggle with aggregation. Interestingly, the MS consultants gave advice that was diametrically opposed to that given by NCR.
The client also uses SAP and CRM (Ceres). It would not be possible to implement a single schema that gives each of these applications the data structure that makes it perform optimally. With 3NF (or thereabouts) as the starting position we can at least provide application- specific table/views to achieve this without sacrificing the core database layout.
As the saying goes 'Different strokes for different folks' - there is no 'best'.
It don't subscribe to the viewpoint that any one approach is more or less performant. Teradata reads tables and performs redistributions on a given platform at the same speed no matter what approach to schema design is employed. If you give it more work to do to answer a question of course the query will take longer. Our job as designers/developers is to ensure that all applications can be used to arrive at the correct answer by performing as little work as possible.
I can't help but feeling that Neil Raden's view that analysts such as Gartner et al simply re-hash the outpourings from vendors is a bit harsh, to say the least. IMHO, NCR are guilty of seriously under-marketing the capabilities of the Teradata product, at least over here in the UK/Europe. I know folks in DW on other platforms who have never even heard of it. Anyone who believes the hype from certain other DBMS vendors (starts with an 'O') marketing department would be naive at best.
I believe that the wholesale move towards tools that generate SQL in response to user input via a GUI (SAS, BO, MS, Ceres etc) is partly responsible for the 'which schema approach should we adopt' debate. They all have different needs/shortcomings/strengths that can't easily be satisfied, if at all, by a single database schema no matter what approach is adopted.
|Copyright 2016 - All Rights Reserved|
|Last Modified: 28 Jun 2020|