IBM Cognos User Manual page 143

Adding New Data to Time-based Partitioned Cubes
New categories are added to the newer cubes in the normal manner. However,
older, pre-existing categories may not have new data associated with them in the
more recent cubes. In such cases, not every child cube may contain the complete
set of metadata; that is, the category information, relationships, and measure
values.
If the category lists do not remain stable over time, or if major changes occur
between cube updates, it may be difficult for Cognos Transformer to understand
and correctly interpret the context of any moved categories. That is, the integrity of
the category hierarchy, or parent-child and sibling relationships, may not be
reflected in each child cube.
Use of the Unique and Move options further complicates the situation, because the
moved category becomes separated from the parent and siblings that indicate its
context. This can produce unexpected sorting results.
For example, suppose your model design includes several child cubes, and the
data added to your time-based partitioned cube includes new categories that did
not previously exist. Because there is an overlap in the time periods, the overall
cube sort order reflects a merging of the categories. More recent categories are
appended to the end of the list, destroying the sort order. The parent context is
unavailable in the child cubes, so the category list cannot include categories
missing from the child cubes. If your query spans multiple child cubes, the overall
sort order is a merging of the categories from the individual cubes.
Example - Sorting After the Addition of Categories to Disjoint
Child Cubes
Suppose your model design includes disjoint child cubes. The overall category
order in the parent cube is A, B, C, D. Categories A and B appear in one child
cube; C and D appear in the other.
Procedure
You add a new category, E, to the A+B child cube. A special algorithm correctly
sets a sort order in the reporting component to A, B, C, D, E, recognizing that E is
the child of the same parent as A, B, C, and D.
Note: Time-based partitioned cubes are optimized to undergo incremental updates
in which the metadata changes gradually over time. They are not designed for
cases where significant numbers of child categories are deleted or moved, using
Move option on the Level property sheet, for example. Operations such as these
remove historical context, producing disjoint child cubes in which unexpected sort
orders may appear.
Slowly Changing Dimensions
In your OLAP reporting components, a group of time-based partitioned cubes
appears as a single virtual cube, but the data in the individual cubes represents
business information that may be changing slowly over time.
Suppose that, over several months, you need to delete a category, add a new
category in its place, and move the added category to a new parent in the same
dimension. Cognos Transformer can maintain the required category code
identifications because the changes are sufficiently gradual, from one update to the
next. Because the parent-child context is preserved, the original sort order can be
successfully maintained.
129
Chapter 6. Creating PowerCubes