KB441987: Best Practices for increasing performance of publishing large Intelligent Cubes

CONTENT:

• Enable cube partitioning, the partition attribute should be the attribute with more cardinality in the cube (more cardinality means the attribute with more different elements, for example in the Tutorial project the attribute Item or Customer).
  Additionally, the number of partitions is based on the number of CPUs in the I-Server. It should always be the amount of CPUs divided by 2, for instance if there are 16 CPUs, the partitions should be 8.  More information about cube partitioning can be found here: https://www2.microstrategy.com/producthelp/10.8/InMemoryAnalytics/WebHelp/Lang_1033/Content/InMemoryAnalysis/Partitioning_cubes.htm  

• Under the VLDB settings for that cube, make sure that to increase the setting "Query Optimizations - Maximum Parallel Queries Per Report" to the same value that is determined in the partitions. For instance, if it is determined there are 8 partitions in the cube, then the value of this VLDB setting should be 8.

• Under VLDB settings > Query Optimizations > Data Population for Intelligent Cubes admins need to test all the following options:
  - Normalize Intelligent Cube data in the database.
  - Normalize Intelligent Cube data in the database using relationship tables.
  - Direct loading of dimensional data and filtered fact data.
  For this point admins need to try the three options and see which one has better performance.

 
Note:
Make these modifications and re-publish the cube to find out which combination gives better performance. Once the best combination is determined, admins can test the deletion and see if the performance improves there.
KB441987