I'm building an SSAS Tabular Cube and have a modeling questions (my example here is simplified).
Let's say I have dimensional data for Region and Warehouse, where one region can contain several warehouses. Then I have 2 fact tables with different grain; Sales facts is on Region-level and Inventory facts are on Warehouse-level. Now I can see 2 different ways of building the data-model (see image):
A. Region and Warehouse are put in 2 different tables and a relationship between them defines their 1-n hierarchy. Each table then has a relationship to their facts.
B. Region and Warehouse are put in the same table (called Location). This table has some rows on Region-level (where Warehouse column is empty) and some rows on Warehouse-level. This enables the different fact tables to relate the same key.
I see that there are some pros/cons with each approach, but can't decide on which is better.
Pros with A:
- Easier to develop and maintain.
- If relationship between Sales and Region is made bidirectional then Sales facts can filter out warehouses.
Pros with B:
- Faster performance because less relationships and keys?
- All location-relevant dimensions in the same table.
- Fewer tables to maintain.
What approach would you say is better and why?
