Aggregation Design and Usage-Based Optimization - Microsoft SQL Server 2008 Analysis Services - page 422

</Attributes>
</Dimension>
</Dimensions>
<Aggregations>
<Aggregation>
<ID>Aggregation 1</ID>
<Name>Aggregation 1</Name>
<Dimensions>
....
</AggregationDesign>
After you create an aggregation design object and set the AggregationDesignID property
of the partition, you create physical aggregations for partition by initiating the processing
of the partition. (Execution of the ProcessFull command, for example, would create
aggregations.)
Partition aggregations are created during the last stage of partition processing (index
processing). If you already have data loaded into the partition, you can change or add
aggregations to your aggregation design object and use the ProcessIndex command to
process the aggregations separately without reprocessing the data. You can also use
ProcessClearIndexes to drop indexes and aggregations from your parturition.
NOTE
Changing the AggregationDesign object or the AggregationDesignID property of the
partition will not cause the partition to drop the existing set of physical aggregations.
You get into a situation where the metadata differs from the partition data. The set of
aggregations defined for the partition differs from the set of aggregations processed for
it. You can use the LastProcessed and LastSchemaUpdate partition properties along
with the LastSchemaUpdate property of the AggregationDesign object to determine
whether the logical aggregation design is out of sync with the physical aggregations.
The Aggregation Design Algorithm
You don't have to design aggregations for your cube manually. Analysis Services provides
the Aggregation Design Wizard, which will design a set of aggregations for your cube. You
can run the Aggregation Design Wizard either from Business Intelligence Development
Studio (BI Dev Studio) or from SQL Server Management Studio.
To make decisions about the best set of aggregations, the aggregation design algorithm
must have an estimated count of the members in all the attributes in all the dimensions
in the measure group. If the EstimatedCount property of a dimension attribute isn't set,
the Aggregation Design Wizard tries to count the members in the attribute by submitting
a query to the relational database (such as select distinct count... ). If your dimen-
sions are large and have many attributes, the wizard's estimation phase can take a while.
After the estimates are obtained from the relational database, the Aggregation Design
Wizard updates the EstimatedCount property of an attribute object. The next time you run