Why can the storage mode of dimension and cube be different?

Question

I understand that MOLAP is to organize and pre-calculate your data in a multi-dimensional structure. Each dimension is an axis, and the aggregated measures are stored in each little space. ROLAP is to query data directly from the underlying relational database.

As the title described, the user can at the same time set the dimension storage mode as MOLAP while the cube's storage mode as ROLAP. How does SSAS handle this situation?

If the aggregation(pre-calculation) only contains dimension data but not measures from the fact table, how does it improve the querying performance?

Answer 1

if your dim data is already in memory, including the primary keys, then you don’t need to query dims to determine parents, descendants, populate levels, etc.

Also, queries that retrieve only a handful of dim keys from the fact table can be performed in a select from the fact table only, without the need to join to the dimension. You filter directly on the fk values.

It won’t solve every performance problem, but will greatly simplify many common queries where you would otherwise need to join with several dims. Even reducing the number of joins in 1 or 2 can have an impact.