Physical Data Warehouse Design - Data Warehouse Systems: Design and Implementation - page 265

The data for a partition in a measure group must be exclusive of the data
in any other partition in the measure group; otherwise, these data would be
considered more than once. Every measure group has at least one partition,
created when the measure group is defined. This initial partition is based on a
single fact table in the data source view of the cube. When there are multiple
partitions for a measure group, each partition can reference a different table
in either the data source view or in the underlying relational data source for
the cube. Also, more than one partition in a measure group can reference the
same table.
Analogously to what we explained in the previous section for database
tables, partitions allow large data cubes to be managed eciently, for
example, by distributing source and aggregate data of a cube across
multiple hard disks and multiple servers. This improves query performance,
load performance, and eciency of cube maintenance. For example, if
in the Northwind data cube we partition data by year, only the last
partition will be processed when current information is added to the
cube. Partitions can later be merged. For example, at the end of a
year, the quarterly partitions can be merged into a single partition
for the year and a new partition created for the first quarter of the
new year. Thus, partitions can be configured, added, or dropped by
the database administrator. Each partition is stored in a separate set
of files. Aggregate data of each partition can be stored on either the
instance of Analysis Services where the partition is defined or on another
instance.
Finally, the storage mode of each partition can be configured independently
of other partitions in the measure group, for example, using any combination
of source data location, storage mode, and aggregation design. We study this
feature next.
ROLAP Storage
In the ROLAP storage mode, the aggregations of a partition are stored
in indexed views in the relational database specified as the data source of
the partition. The indexed views in the data source are accessed to answer
queries. In the ROLAP storage, the query response time and the processing
time are generally slower than with the MOLAP or HOLAP storage modes
(see below). However, ROLAP enables users to view data in real time and can
save storage space when working with large data sets that are infrequently
queried, such as purely historical data.
If a partition uses the ROLAP storage mode and its source data are
stored in SQL Server, Analysis Services tries to create indexed views to store
aggregations. When these views cannot be created, aggregation tables are not
created. Indexed views for aggregations can be created if several conditions