About good practice in FrameWork

[jive]

Hi everyone,

I have two questions, I almost read all posts in the good practice for framework manager and documentation too, and I'm not an expert believe me. I understand most of time people work with data not exactly build in star model in the datawarehouse,database, side. In the project I work, this is not the case, we actually put a lot of work on the database side of this project. The way we think is if the source of the data are already define in star schema it will be faster for Cognos. We also think the physical layer in the good practice, first layer, can have the relationship because everything is already define with referential integrity in the database, reference key and all the proper indexes. We tried to build a perfect data source for Cognos to work with. Did we think right to add the relationship on the physical layer in FrameWork Manager, it's still confuse for me.

Second question, we only have one star schema per business subject , we also only have one presentation layer were the dimension are defined. Is it okay to do that? I ask that because in good practice the dimension layer is separate from the presentation layer.The number of dimension are around 30 to 40 for each business subject with some going up close to 100. I Know that number is high but we have to give what the client need and ask for, even if they are warn that report could be slower due to the number of dimension and the volume of the package.

Thanks for your advice ;-)

[blom0344]

'Tuning' the database to reporting needs will make it possible for creating a near ideal Cognos model with joins in the database layer and just one model layer on top of that. Actually if you are not going to build cubes from the model , but stick to DMR for dimensional modelling I think the number of dimensions is fairly academic. Cognos will only generate what is asked when defining an analysis or report. The size of the package itself is not relevant, perhaps a larger package may take a little more time to load into memory, but performance is dictated by what data needs to be handled.

Do you need to accomodate for multi-fact analysis, by the way?

[jive]

Hi Blom

for the fact the answer is no but we have 10 measure in the same table, but like I said before the have all the same granularity and they are all related to the same business subject.

Thanks for your answer  and that's confirm what I think about the multiple layer, in the project I'm working. On other project this will be a total different way of thinking.