Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 303

In order to be able to query a cube, we also need to run a Process Structure on
it, which usually takes just a few seconds. This can be run the first time a cube is
deployed to production and will only be necessary afterwards if the cube's structure
is changed. If we do this and do not process any partitions, the cube can be queried
but will contain no data; we can then process partitions individually, and as each
partition finishes processing, its data will appear in the cube. This behavior can be
useful in a development environment if we want an exact copy of our production
cube but don't want to load all of the data into it. In this case, we can run a Process
Structure on the cube and then a Process Full on just one or two partitions.
Processing an object will lead to the Storage Engine caches related to that
object and the entire Formula Engine cache being cleared: if the data in
the cube has changed, any data in the cache may no longer be valid. This
is especially important if we have to perform any kind of processing
during the day because query performance will suffer as a result, and
the more often we process, the less benefit we will get from caching.
Lazy Aggregations
Lazy Aggregations is a processing option for partitions. Normally, when we run
a Process Full on a partition, it will be available when both the Process Data and
Process Index steps have completed. If we want the partition to be available sooner,
we can set the Processing Mode property on the partition to Lazy Aggregations .
This means that the partition becomes available as soon as the Process Data step has
completed; the Process Index step will then be run in the background as and when
resources are available. The price we pay for having the partition available earlier is
that, while aggregations and indexes are not built, queries against the partition will
perform worse.
Using the Lazy Aggregations option, we lose control over when the aggregations
and indexes for a specific partition will be built because it will depend upon several
factors, most of which are out of our control. Also, Lazy Aggregations interfere with
our ability to synchronize a cube, as we will see later. As a result of these drawbacks,
we do not recommend the use of this option.