FM package testing expectation

[sdf]

Hi,

Just recently I created a package (relational).
This package is composed of query subjects that have 7-10 tables each. The query subjects are not inter-related.
So I am testing each query subject by creating a report from each.
I know the performance of the queries or how the report would run depends on how good the package was modeled.
But  I want to know in general if what I'm experiencing in my unit test is somehow normal.
So I created a list in COgnos BI, I selected 1 data item each from one query subject that have 10 tables.
The run time takes more than two hours.
One report sometimes finishes, sometimes just error after hours of runtime.

Is this expected?
Any tips on making my unit test easier?

thanks!

[Lynn]

Hi,

Just recently I created a package (relational).
This package is composed of query subjects that have 7-10 tables each. The query subjects are not inter-related.
So I am testing each query subject by creating a report from each.
I know the performance of the queries or how the report would run depends on how good the package was modeled.
But  I want to know in general if what I'm experiencing in my unit test is somehow normal.
So I created a list in COgnos BI, I selected 1 data item each from one query subject that have 10 tables.
The run time takes more than two hours.
One report sometimes finishes, sometimes just error after hours of runtime.

Is this expected?
Any tips on making my unit test easier?

thanks!

A package would generally be comprised of a fact query subject (or multiple fact query subjects) connected to multiple dimension query subjects. A single query subject made up from 7-10 tables doesn't sound like a star schema which is what Cognos (and indeed many BI tools) typically expect. Not sure if I am interpreting what you described correctly or not.

The FM user guide includes guidelines for modeling relational sources and covers star schema concepts, cardinality for fact identification, resolving traps, and much more. Certainly you should look at the SQL being generated by various scenarios to assess if it is doing as expected or not.

If you could provide more information about what your tables represent, how they are joined, and/or examples it might help you get better advice on how to model properly. Two hour run times and errors after hours of running does not sound like an expected outcome.

[sdf]

thanks for that.

Well, there's not really a modelling data that happened. It was just a matter of importing the tables from the DB then creating relationships from the imported tables. We did not create any materialized view or new tables. There's no defined fact table. Most of the table holds their own measures.

:(