A few questions about Decision Stream!!

[srikalyan]

Hello,
I have a few very basic questions:

1.) How would be the landscape for a DS project? Would it be something like the traditional Development, Quality and Production system landscape?

2.) Do you typically require a data staging area while using Decision Stream? What I mean is do we pull the data from the source systems, then store them in a database or data staging area and then from this staging area we do the transformations ans push it to the target database systems or datamarts? Or is it that we just map the source systems table fields to the target system fields directly with a transformation model in between?

3.) Are there any extracting \ loading strategies\practices?

4.) How do we conduct the quality assurance or testing in these kind of ETL projects generally? Any particular testing strategy or practice?

5.) How do we do error handling in these kind of projects?

Many thanks in advance!!!

[MFGF]

1) Yes - this would be a normal scenario for most development environments.

2) It's normally best practice to assemble all the data in one place in some form of staging area before beginning to go through the transformation process.  DecisionStream will allow your transformation builds to read directly from the source systems if that's what you want to do, but normally you'd want to get the data off these as fast as possible so they can get back online with minimal interruption, so a quick 'data transfer' build into the staging area would achieve this in the shortest time.

3) Cognos tends to follow Ralph Kimball's ideas on data warehousing, and DecisionStream is designed with these principles in mind.  Take a look at his books for lots of good ideas on strategies and practices.

4) Just like you would in any development environment.  The builds need to be tested by the developers creating them to make sure they are defined correctly, and user acceptance testing needs to be done to ensure the design is correct and you are delivering what the business requires.

5) There is lots of inbuilt error-handling technology in DecisionStream.  Bad dimensional data can be allowed or rejected where it gets assembled into the Dimensional Framework hierarchies, and there are various logging options to allow you to see what bad data is being detected.  Bad transactional data can be allowed or rejected too, and again there are various options to allow you to identify the bad transactional data - either in a reject file or delivered to separate relational tables in a database.

Hope that helps!

MF.

[srikalyan]

Thanks. That was helpful.