Database Reference
In-Depth Information
•
If existing members do change but we don't mind waiting for these changes
to appear in the Analysis Services dimension, we can run regular Process
Adds and then run a Process Update or even a Process Full on a less regular
basis. This can be useful if we need to add new members to a dimension
during the day with minimal impact to our users, but can wait until our
nightly processing window for our dimensions to be completely updated.
For very large dimensions consisting of several million members or more than of
attributes, splitting a Process Full up into separate Process Data and Process Index
operations may allow us to make better use of available memory. Whether it
does improve processing performance or not will depend on the structure of the
dimension and the server's hardware; thorough only testing is will determine
if it is necessary to determine whether it is worth doing this. Similarly, as
mentioned in
Chapter 3
,
Designing More Complex Dimensions
, setting up the
AttributeHierarchyOptimizedState
and
AttributeHierarchyOrdered
properties to
False
may also improve processing performance at the expense of,
respectively, query performance and member ordering.
Last of all, remember that many structural changes to a dimension (for example,
adding a new hierarchy) will mean that a Process Full is necessary. As a result, such
structural changes will need to be carefully planned for.
Partition processing
As with dimensions, partitions can be processed in several ways. Processing a large
partition is a very CPU intensive operation, and as a result, partition processing
normally takes longer than dimension processing. We therefore have to think more
carefully about how to process our partitions in the most efficient way. Luckily,
partition processing is also a lot less complex than dimension processing, in that
processing a partition has no side effects on any other objects.
The following table lists the different options for processing a partition:
Type
Description
Full
The partition is completely rebuilt. All data inside the partitions is deleted
and reloaded from the relational database; indexes and aggregations are
rebuilt. It is equivalent to an unprocess, followed by a Process Data and a
subsequent Process Index.
Search WWH ::
Custom Search