After having explained Bill Inmon’s point of view on what the premises and foundations should be when building a datawarehouse, it is now time to study Ralph Kimball’s approach, a somewhat different view from start to finish.
Unlike Inmon, Kimball defends a “Bottom-up” work methodology. By this it means that the procedure to follow to build a datawarehouse is to start initially with small components to evolve to higher structures and models. And this is so because for Kimball a datawarehouse is nothing more than the union of the different datamarts of an organization.
The philosophy of the Kimball approach
Its philosophy focuses on the fact that, in most organizations, the construction of a datawarehouse originates from the interest and effort of a department. This is why in its first version this datawarehouse is nothing more than a departmental datamart.
As other departments need their own datamarts, they will be combined with the first one maintaining a standardization methodology through what Kimball calls “shaped dimensions”, which will be the common dimensions between the different departments. The key is that these dimensions must be shared by the different datamarts that exist in the organization, thus guaranteeing their integrity and giving rise to the conglomerate of structures that make up the datawarehouse for Kimball.
To achieve this result, it is important that these shaped dimensions have a consistent and suitable design for all datamarts, so that when creating a new one, reuse the already defined dimensions, whether or not to include other new dimensions.
The main advantage of this data warehouse approach is that, being formed by small datamarts structured in dimensional data models (star or snowflake schemes), specially designed for consultation and reporting, the entire datawarehouse can be exploited directly by reporting tools and data analysis without the need for intermediate structures.
Arquitectura Kimball – modelo dimensional
Regarding the questions about granularity, although this type of datawarehouse usually presents the aggregated data based on the queries and reports that must be generated, Kimball insists on the need for these aggregations to be complemented with higher level data of detail.
The argument is that the business questions that users may ask are unpredictable, so that the datawarehouse must be prepared to answer all of them, guaranteeing data exploration and navigation through hierarchies from aggregated data. to disaggregated information.
Kimball calls this type of architecture as “Data Warehouse Bus Architecture” and the four fundamental steps that must be followed to build this type of database are, first of all, the identification of the business process to be studied, the definition of the granularity of the data, the selection of the dimensions and attributes and, finally, the identification of the facts or metrics.
The architectural scheme based on the foundations of Ralph Kimball would be the following image:
As you can see, both Kimball and Inmon share the need to establish an integrated and stable data storage system that guarantees the exploitation of information, responding to all business questions that may arise. However, their philosophies when building a datawarehouse differ greatly from each other, and it is not easy to argue which of them is the most valid. Despite this, in the next post I will make a summary with a comparison of both visions trying to explain in which situations a model can be more reliable than the other.