Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 302

• Process Add is normally only used with the Standard Edition of Analysis
Services, where we are only allowed to have one partition per measure
group, and so strategies like the one described in the previous bullet are not
possible. With only one partition to store all of our data, a Process Full will
involve reloading all of the data from our fact table whether it is new or not,
and this could take a long time. Using Process Add to load only new rows is
going to take much less time, even if we still need to schedule an occasional
Process Full as noted earlier.
• If our partitions are very large, processing data and building aggregations
might require a great deal of memory. It might be better to perform a Process
Data in parallel for all the partitions that need to be processed, and then
when this has finished and the memory has been released, we can run a
Process Index on them. In this way, we can separate the two processing steps
and reduce memory pressure on the server while also having greater control
over when load is placed on the relational database.
• It is very difficult to say whether this approach will benefit processing
performance - thorough testing is necessary. If memory is not an issue and
we have a powerful server, then processing both data and indexes at the
same time might increase parallelism because when the CPUs are waiting
on the I/O operations involved in a Process Data, they can be used to build
aggregations and indexes instead.

• If dimensions have been updated using Process Update, some aggregations

on existing processed partitions may no longer be valid and will need to be
rebuilt. In this case, running a Process Default on all these partitions will
rebuild only the invalid aggregations. Therefore, if we are using Process
Update and loading data into newly created partitions at the same time, our
workflow needs to be as follows:
1. Run a Process Update on all dimensions that need it.

2. Build any new partitions that are necessary and perform any other

partition management tasks.

3. Run a Process Default on all the partitions in the cube. This will have

the same effect as running a Process Full on the new partitions, and
rebuild any invalid aggregations on the existing partitions.