Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 299

Type Description

Update Compares the contents of the relational dimension table with the contents

of the Analysis Services dimension and then deletes any members from the
Analysis Services dimension that no longer exist in the relational table. It
also updates any members that have changed and adds any new members.
Since existing members on hierarchies with at least one flexible attribute
relationship in the chain of attribute relationships down to the key attribute
can change, then all aggregations including these hierarchies will be
dropped. They can be rebuilt later by running a Process Default on affected
cubes or measure groups.
Over time, the performance of a dimension will degrade after multiple
Process Updates. As a result, if do you use Process Update regularly, it's a
good idea to schedule an occasional Process Full, for example, at a weekend
or public holiday when it will cause minimal disruption.

Unprocess Deletes all structures and data for the dimension.

Default Performs whatever processing is necessary to bring the dimension up to a

fully processed state.

With these options in mind, here, in order of increasing complexity, are some of the
possible strategies we can use for processing dimensions:

• We can run a Process Full on all of our dimensions, which of course means

we then have to run a Process Full on all of our cubes. As we have already
said, this is the simplest option but is only feasible if we have enough time to
do it - we need to be sure that we will have enough time in the future as the
data volumes in our data warehouse grow. Remember that adding hardware
can have a dramatic impact on processing performance, and may be cheaper
than developing and maintaining a more complex processing solution.

• We can run a Process Update on any dimensions that have changed

their contents, and then run a Process Default to rebuild any invalidated
aggregations. Note that Process Update can take a significant amount of
time to complete on large dimensions.

• If existing members on dimensions never change, we can run a Process Add

to add any new members that appear. Although Process Add can be much
more difficult to configure than Process Update, it can perform significantly
better. However, it can only be used in certain circumstances. For example, in
scenarios where we are only making Type 2 changes to a dimension, it is safe
to use. The fact that Process Add does not invalidate existing partition data
or aggregations is also very important as it makes it very performant because
partitions and aggregations do not need to be reprocessed or rebuilt.