Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 301

Type Description

Add Also known as incremental processing, this adds new data to the partition.

After we've specified a table or SQL query containing new fact rows,
Analysis Services creates a new partition, fully processes it, and then merges
the new partition with the existing one.
Over time, the performance of a partition will degrade after multiple Process
Adds. Once again, it's a good idea to schedule an occasional Process Full if
we are using Process Add regularly.

Data Behaves the same as Process Full but does not process indexes and

aggregations.

Index Builds indexes and aggregations for the partition. This is normally used after

a Process Data.

Default Does whatever processing is necessary to bring the partition up to a fully

processed state. If the partition itself is in an unprocessed state, a Process
Full is performed. If, on the other hand, the partition is processed but
some indexes or aggregations inside of it are not, only those indexes and
aggregations will be processed.

Unprocess Deletes all structures, data, indexes, and aggregations from the partition.

There are a different set of issues to consider when choosing a partition processing
strategy compared to a dimension processing strategy. Dimension data changes
naturally, and these changes are something we have to plan for. On the other
hand, existing data in our fact tables does not usually change over time. New rows
will appear in fact tables and recently added rows may be updated. Unless older
data is incorrect or we need to recalculate measure values based on new business
requirements, this older data is normally static.

With these considerations in mind, let's take a look at some scenarios where the
different partition processing options can be used:

• For new partitions, or where data in existing partitions needs to be

completely reloaded, we have to do a Process Full. The important thing here
is to set the slices for our partitions so that when we do need to do a Process
Full, we do the minimum amount of processing necessary. For example, if
new data is loaded into our fact tables every month, every month we can
create a new Analysis Services partition to hold that new month's data and
run a Process Full on just that new partition. Assuming that the data for
previous months does not change, no other partitions need to be processed.