IBM Cognos User Manual page 198

12. Open the cube in your OLAP reporting component and try drilling down to
Example - Partitioning Manually When Auto-Partitioning Cannot Be Used:
Partitioning can be difficult. If your manual optimization efforts are not effective,
you may have to ask an expert for assistance.
Suppose you have a flat-dimensioned model consisting of product IDs, customer
numbers, and countries or regions. The ideal parent:child ratio of 1:10 is exceeded;
there is only one parent for each of the large child category levels (Product Brand,
Customer Type, and Region).
Your customers need summary reports on categories from the Product and
Customer dimensions, so you manually partition at a higher level in those
dimensions. You specify which levels and categories are included in the
partitioning and set the Maximum number of passes on the Auto-partition tab to
that number of partition levels plus one.
However on testing, you discover that additional partitions are needed. You try
various strategies, using the Products dimension.
Procedure
1. First, you assign a single partition to the entire Product Type level, which
2. To more evenly distribute the total number of categories, you navigate to the
3. You rebuild the cube, check the partition status, and confirm that the Product
4. You decide to add individual Country or Region categories to the same
184
IBM Cognos Transformer Version 10.1.1: User Guide
window shows the dimension name, the category code, and its record count.
Check to see if any of the partitions are larger than your specified partition
size.
the partition with the largest number of records. If performance is
unacceptable, reset your partitions and add another level of partitioning to the
dimension you chose. Alternatively, select a different dimension where the
gains may be more favorable. For each new level of partitioning, remember to
increase the partition number by one (1).
results in the following:
Level 0, All dimensions, <Summary> category – record count of 237
Level 1, Line dimension, <Type 1> category – record count of 31
Level 1, Line dimension, < Type 2> category – record count of 181
Level 1, Line dimension, < Type 3> category – record count of 25
However, drill-down performance is only acceptable when querying Types 1
and 2. It is unacceptably slow for Type 3, because there are too many records in
the partition.
General tab of the Category property sheet for each product type, and assign
the same partition number (1) to all Type 3 child categories, and to all Type 1
and 2 categories at the parent level.
Type category counts are better balanced, with no partitions containing more
than 87 records. However, you note that some Region categories contain a large
number of child records.
partition as that assigned to the parent, to further optimize partitioning. You
then recheck query performance, but conclude it is still unsatisfactory.