Using ROLAP (Relational Online Analytical Processing) or MOLAP (Multidimensional OnLine Analytical Processing) is something to be raised at the stage of physical design and affects how the tool of exploitation of the Data Warehouse 'attack' to the data .

If ROLAP, the attack tool to exploit the information directly to the stars that have been created in the relational DBMS chosen.

If you use MOLAP technology, although we have created these stars, the tool will create their own multidimensional cubes optimized for the engine to implement. In this case there are enough of implementation options: you can make it go directly from an ODS (Operational Data Store) to build MOLAP cubes, and also keep the stars in the relationship, or even create the cubes directly, bypassing the relational. 
I would also keep the whole environment implemented on relational, rather ligarnos and the technology implemented by each tool for the creation of buckets.

For example, SQLServer Analysis Services with the implementation of ROLAP cubes, MOLAP HOLAP or is as easy as dialing a radio button to define the cube, the tool makes it transparent, so it can be as simple as what to study option is best depending on the size and forecast growth and utilization of DWH, do some test and choose the one that meets our needs.

However, most tools is that prefer the use of one technology or another. For example, DSS (Decision Support System) Microstrategy opts for the use of ROLAP with a relational DBMS, and SAP BW or Essbase Analytics (Hyperion), always implement their own multidimensional cubes.