Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 297

Analysis Services processing can be broken down into two different tasks:

• Dimension Processing : This involves loading data into a dimension and

building indexes on it.

• Partition Processing : This is more complex. Before we can query a cube, all

of the dimensions in the cube need to be processed, and we need to process
the cube itself too. A cube is made up of measure groups, and a measure
group is made up of partitions. Since partitions are where the data in a cube is
actually stored, then when we talk about processing a cube, what we are really
talking about is processing all of the partitions in a cube. Processing a partition
involves loading data into it, building indexes, and building any aggregations
specified in the aggregation design associated with this partition.

MOLAP, HOLAP, or ROLAP?
So far, we've assumed that we're always going to use MOLAP storage
mode for all of our partitions. That's because 99 percent of the time
we will be! MOLAP storage involves storing all data in Analysis
Services' own data structures; it gives us the fastest query performance
but involves the longest processing times. ROLAP storage leaves all
data in the relational database, and when MDX queries are run, then
Analysis Services generates SQL statements to retrieve the data it
needs; querying is relatively slow as a result, but processing is very
fast. HOLAP storage is a cross between the two and involves storing
aggregations and indexes in MOLAP mode but everything else in
ROLAP mode.
ROLAP storage for partitions is only very rarely useful, for when we
really do need to see real-time data inside a cube. Even then, the poor
query performance of ROLAP partitions means it can only be used
with relatively small data volumes. Since fast query performance is
the number one priority for any Analysis Services cube, the extra time
needed for processing MOLAP partitions is usually a price worth
paying. We have never seen a scenario where HOLAP storage is the
right choice, so we can safely ignore it.
Similarly, MOLAP storage is almost always the right choice for
dimensions. As we saw in Chapter 4 , Measures and Measure Groups ,
ROLAP storage is usually only used for very large fact dimensions
that would take too long to process in MOLAP mode.