Dimension and Partition Processing - Microsoft SQL Server 2008 Analysis Services - page 383

Full Dimension Processing
Fully processing a dimension (the ProcessFull processing option) causes it to drop all
data structures associated with that dimension. New data is loaded into the dimension
from the dimension tables and all dimension structures are rebuilt. Analysis Services
assigns a new metadata version to the dimension. Data in the partitions of all measure
groups that include this dimension as a regular dimension is deleted—you must reprocess
these partitions.
Incremental Dimension Processing
Incremental processing is typically the most efficient way to update a dimension. You can
only add new members to a dimension during incremental processing (the ProcessAdd
processing option). The existing members in the dimension retain their DataID values,
and new members added during incremental processing receive new DataID values. You
can make the process operation load data from the original relational table or view that
contains both old and new members, or you can specify a new table or view that contains
only new members. If you use a table that contains both new and old members, Analysis
Services updates the properties of existing members (except the DataID values and the
DataID values of related attributes) and adds new members to the dimension. During
incremental processing, Analysis Services must rebuild quite a few data structures. It has to
rebuild hierarchies because adding new members to the dimension can change the order
of members and decoding tables. However, the situation is different for indexes. Analysis
Services tries to save processing time by rebuilding indexes only for segments that contain
new members (in other words, the last segments of the attributes).
The biggest advantage of using the ProcessAdd option is that partitions and partition
aggregations are not invalidated. No additional cube data structures are required to be
rebuilt because the DataID values of existing members used in the fact data, and any exist-
ing relationships between such members, remain the same. The new members added to
the dimension don't have fact data associated with them in the partitions yet, so adding
the new members does not invalidate the existing fact data.
Updating Dimension Processing
When you specify the ProcessUpdate processing option, Analysis Services changes the
data in the dimension according to changes of the data in the relational data source. This
processing option differs from the ProcessAdd option because Analysis Services requires
that the relational table or view must contain all the members, both old and new, to be
included in the dimension. Any member not found in the relational table or view is
considered deleted.
While updating a dimension, you can modify any property of a member. Because of this
operation, relationships between members can change and dimension members can
appear under a different parent in a hierarchy. For instance, if customer Sasha Berger
moved from Bellevue to Redmond, the City related attribute would change to Redmond .
This is possible if the attribute relationship is defined as Flexible . If the attribute relation-
ship is defined as Rigid , Analysis Services reports an error when it detects such a change.