The Inmon approach >> Kostis Panayotakis

Kimball vs Inmon

In parts 2 & 3 of this article series, we described the data warehouse architecture according to the Kimball and the Inmon approach. In the present article we shall describe the main differences between the two approaches and their common points.

The two approaches have the following common points: o The proposed use of a staging area, when the volume of data and the extraction-transformation-loading (ETL) complexity is high o The implementation of automated ETL processes o The use of multidimensional structures and analysis at the data mart level, based on the dimensional model and on-line analytical processing (OLAP) tools o The use of an iterative development approach, which is however based on different design and development methodologies. The main differences are identified at the following points (K for Kimball, I for Inmon):

Data warehouse development philosophy K: Based on the prioritized selection of specific business processes. I: Based on the Enterprise ‘data model’ as it is defined by this approach.

K: Direct development of data marts on the selected business processes. Exclusive use of denormalized dimensional models (star schemas). I: Development of the Enterprise Datawarehouse (EDW) based on a normalized database schema. The development of data marts, is based on data retrieved from the EDW.

Data mart definition K: A data mart maintains data of the lowest level of detail (atomic data), which relate to a business process. Data marts are developed based on the popular dimensional modelling methodology. I: A data mart maintains aggregate data which relate to a Business Unit. They are built to monitor predefined KPIs (key performance indicators).

K: A data mart is built by extracting data directly from operational systems. I: A data mart is built by extracting data from the Enterprise Datawarehouse (also called dependent datamart).

K: Data marts are linked to each other, based on conformed dimensions. I: Data marts are not linked to each other.

K: A data mart maintains all available historical data. I: A data mart maintains limited history, since history is kept in the Enterprise Datawarehouse.

Phased development approach K: phased development of datamarts on selected business processes, which are linked on conformed dimensions, forming the datawarehouse Bus architecture. I: design of the whole Enterprise Datawarehouse based on the Enterprise ‘data model’. Phased implementation of subject areas, according to priorities set.

International experience records difficulties in the successful implementation of the Inmon approach. On the other hand, enterprises which have developed independent, incompatible and uncoupled data marts without central coordination, are facing the challenge to consolidate them, in order to yield combined data analysis value. Consolidation requires redesign of a major part of the existing infrastructures. The Kimball approach, which receives increasing attention, does not propose implementation of uncoupled data marts.