Database Reference
In-Depth Information
srv.ini
as well. Reference
Chapter 8
,
Administering and Monitoring Ana-
lysis Services
, for additional information about configuration options.
• If you see any temporary files created during the aggregation processing,
your server does not have sufficient memory. Consider adding memory or
separating querying and processing activity onto dedicated SSAS instances.
Temporary files can be found under the
<TempDir>
folder, as specified in
the
msmdsrv.ini
configuration file.
• Experiment with various degrees of parallelism to see which one works best
in your environment. If you have dozens (or even hundreds) of partitions pro-
cessing, all of them in parallel might not work well, because the relational
source might not be able to handle this volume of queries in parallel.
• Watch out for conflicting jobs, for example, processing and synchronization
stepping on each other.
• Remember that you have multiple options for dimension processing. Con-
sider using
ProcessAdd
in lieu of
ProcessUpdate
if your dimension tables
only have new rows added (no updates or deletes).
ProcessUpdate
needs
to check each partition to ensure that none of the indexes have been invalid-
ated. It is not uncommon to see
ProcessUpdate
read dimension data very
quickly, but then spend more time refreshing indexes.
• If processing a particular dimension uses an excessive amount of memory
and heavily taxes the relational source, consider switching to the
ByTable
processing group option instead of the default value of
ByAttribute
.
• Use out-of-line bindings if you have gigantic dimensions with millions of
members.
• If your host does not have much memory it could be helpful to separate the
processing partition data from index processing.
• If reading data from the relational source is quick but writing takes a long
time, be sure to check the disk counters. Remember that each SSAS data-
base could consist of a huge number of files. Suboptimal disk subsystem
could hurt both processing and querying performance.
• If you have a large database which includes rarely-queried historical parti-
tions, consider using ROLAP storage mode for such partitions, while using
MOLAP for frequently-queried partitions.
• In rare scenarios, it might be beneficial to experiment with HOLAP. Remem-
ber that HOLAP leaves data in the relational format, but builds aggregations
on the Analysis Services host. If the relational data source does not support
indexed views (that is how ROLAP aggregations are implemented), the only
choice is to build aggregations in a multidimensional format. Keep in mind,
Search WWH ::
Custom Search