The Kimball Method

A dimensional (denormalized) data warehouse is the product of a series of translations.

1. Data Sources are analyzed.
2. A normalized model called the Operational Data Store is created to represent the enterprise.
3. A denormalized Dimensional Data Store is created based on the normalized Operational Data Store.
4. ETL routines are written to move data from the Source Systems to the Operational Data Store
5. ETL routines are written to move data from the Operational Data Store to the Dimensional Data Store
6. Data is then extracted transformed and loaded from the Data Sources into the Operational Data Store
7. Data is then extracted transformed and loaded from the Operational Data Store into the Dimensional Data Store
8. Analytical tools are used against the Dimensional Data Store for reporting
9. The two step ETL process (6, 7 & 8) is repeated on a schedule as frequently as every day

This is the Kimball method.

Similar processes are used to create subsets that function as Data Marts.

I can't comment on Inmon's strategy except that he is in favor of normalized models (Operational Data Stores or even properly normalized Source Systems) being used without modification as processing power increases and makes the performance barriers of multiple joins go down. Kimball's rebuttal is that a denormalized dimensional model is easier to grasp by the end user than a complex normalized data model.
__________________