The cube metadata is always stored in the SSAS server, but as an administrator, you can control the storage locations of the cube data and any aggregations.
MOLAP(Multidimensional OLAP)-Metadata, Aggregation and Data are stored in SSAS. Data is stored in a compressed format on the server. Storage size is 20-25% of the relational data. The administrator can enable proactive caching on a MOLAP partition to implement real-time data refreshing. It’s like a snapshot at a time. The cube is always available for reporting services.
HOLAP(Hybrid OLAP): Metadata and aggregations are stored on the SSAS server and Data is stored in the relational database. HOLAP is the most efficient mode in terms of disk space because detail-level data is not duplicated, as it is with MOLAP, and HOLAP requires less space to store aggregations than ROLAP does.
ROLAP(Relational OLAP): Metadata is stored on the server.
Both the cube data and the cube aggregations remain in the relational database, so the SSAS server must create additional relational tables to hold the cube aggregations.
Actually it stores the aggregations in indexed views in the relational database.
if you need a low-latency partition and you cannot use proactive caching, ROLAP can be a practical solution.
—————————–Proactive caching———————————————————–
With MOLAP, the server brings the cube data from the data source into the cube when the cube is processed. The data is duplicated because it exists in both the data source and the cube. MOLAP data latency is high because new data is available only when the partition is processed. However, the administrator can enable proactive caching on a MOLAP partition to implement real-time data refreshing.
Proactive caching is especially useful when the relational database is transaction oriented and data changes at random.
When data changes are predictable–such as when use and extract, transform, and Load(ETL) process to load data—Consider processing the cube explicitly. When the data source is transaction oriented and you want minimum latency, consider configuring the cube to process automatically by using proactive caching.
