Hi, is there also a part 3?
I was also wondering about the suggestion made to push back filters to the original data source.
This does work in the example you have given, but some questions:
- in part 1 the suggestion is made to make dim tables import and fact tables dq. Would this not by default generate multiple queries as is the case for the measure? Or is dual mode a 'best of both worlds'? Is a pure DQ model in such cases not better than a composite?
- what about calculation groups? How do they affect the performance/architecture?
- our fact table has information per 5 minutes and needs to be analyzed in the combination of date and time (eg they want to see results from 5PM to 8 AM next day). This makes it impossible to split the date and time dimension (as you are not able to combine these in a continuous axis or even let the user select the time frame given as an example). This leads to a large dimensional table (+300,000 rows for 3 years). Any suggestions?
Thank you for the wonderful article!
Microsoft