IBM Cognos User Manual page 19

v You can use Cognos Transformer to presummarize the data when your users do
not require access to all the details in the source.
For example, if your organization processes 50,000 transactions daily, and you
create the cube weekly, you can summarize the transactions at the weekly level
before Cognos Transformer begins processing. This will greatly speed up cube
creation.
v Consolidation, combining records with identical non-measure values, reduces the
size of the cube and improves performance in your reporting application.
Consolidation is enabled by default in Cognos Transformer. Evaluate your data
to see if it can be further consolidated by using the Duplicates rollup or
Regular rollup features of Cognos Transformer.
For consolidation purposes, non-measure values are considered identical if they
meet any of the following criteria for the particular rollup:
– The source data contains transactions with identical non-measure values.
For example, two sales of the same product are made to the same customer
on the same day, but the colors differ. If colors are omitted from a dimension
view using the Suppress or Summarize command on the Diagram menu, the
sales records will have identical non-measure values.
– Records become identical when a dimension is omitted from the cube.
For example, two sales of the same product are made at different stores on
the same day. If the Stores dimension is removed from the model, these sales
records will have identical non-measure values.
– Records become identical because of the Degree of detail setting on the Time
tab of the Column property sheet.
For example, if the Degree of detail is set to Month for a column associated
with a time dimension that includes week and day values, Cognos
Transformer ignores the week and day values in the source transactions when
consolidating records.
v For queries based on relational packages, enabling the Auto summarize feature
on the General tab of the Data Source property sheet also helps reduce the
number of rows that Cognos Transformer retrieves from the source data, further
improving cube build performance.
Separate Your Structural and Transactional Data
Processing time improves when Cognos Transformer can query your structural and
transactional information separately. You must identify which data sources contain
purely structural information, which contain transactional information (measure
values or facts), and which contain a combination of the two.
When processing queries to create a PowerCube, Cognos Transformer orders the
queries, first reading the structural queries and then reading the transactional
queries.
Ideally, you should define each dimension or drill-down path with a separate
structural data source, and then add one or more transactional data sources to
provide the measures for those dimensions. This restructuring exercise helps to
partially normalize your data, speeding up both the category generation and cube
creation stages.
The best approach is to have unique levels near the bottom of the dimensions, and
to have the transactional queries link to the dimensions using those levels. This is
basically the star schema or snowflake method of creating dimensions in a
relational database. This type of design promotes faster processing because each
Chapter 2. Planning Your Model
5