Data Mart Philosophical Design Differences

[dgorbea]

Hello,

I'm currently working on a project that entails developing a data mart to deliver logistics information
through SSAS cubes. My approach has been to first design a physical data mart
independent of the delivery mecahnism. SSAS cubes, SSRS reports, etc.,
will be used to deliver the data. The source system is a combination of OLTP and ODS
databases.

There's another school of thought at my current company: Rather than create a separate
physical data mart, create views over the current OLTP tables as a means of implementing
a "logical data mart", and deploying the SSAS cubes based on this logical view of a data mart.
Is this approach recognized by the Kimball Group?

[BoxesAndLines]

I'm not sure you'll get a response from the Kimball Group. They frequent the boards infrequently. I'll put on my un-official Kimball cap though. Bad idea for all the reasons why data warehouses started in the first place.

[ngalemmo]

Yep. It's been done before and the end result eventually turns into a major headache.

[Mike Honey]

I agree wholeheartedly with the previous two comments.

IMO it's a particularly bad idea when feeding SSAS cubes, as there are very limited options in that tool alone for handling dirty data. Once you start pointing it at live data, this leads to frequent processing failures which are difficult to debug.

I think a solid star schema is definitely a worthwhile investment for the architecture you described.

Good luck!
Mike

[Sponsored content]