* Multidimensional OLAP (MOLAP)
Keeps detailed data and aggregations in a multidimensional compressed proprietary structure.
With this method, the relational source is accessed only during processing,
and queries are resolved entirely from the multidimensional store.
It provides the fastest query performance,
but it requires a periodic refresh to stay consistent with the relational data.
On average, the amount of space required to maintain a copy of detailed data
into the MOLAP store
is about 20 percent more than the space required for the relational source
(without indexes).
The query performance gained by using MOLAP
more than compensates for additional space required.
* Hybrid OLAP (HOLAP)
Keeps aggregations only in the multidimensional store
and leaves detailed data in the relational source.
Processing is faster than MOLAP
because the activities are limited to calculating and storing the aggregations.
Queries that require detailed data are slower than queries
that can be resolved using aggregations.
This option may be best for historical data that is infrequently accessed.
* Relational OLAP (ROLAP)
Keeps detailed data and aggregations in the relational source.
Queries run more slowly as a consequence
and processing runs more slowly when aggregations are involved.
The most common reason to use ROLAP is to support real-time analysis
when the source database is continually updated.
Source:
Microsoft SQL 2005 Administrator's Companion,
Chapter 22 - Analysis Services
Page 697