Presentation of the bottom-up Data Warehouse Model by Ralph Kimball. He is a well-known author and expert in the field of data warehouses. He is a proponent of a bottom-up approach to data warehouse design. This method is called Bottom-Up Design.
As in other methods, the bottom-up approach creates data shops (Datamarts). To provide dedicated reporting and analysis capability for specific business processes. Thus, easier to use than complex data warehouses.
Definition of the Ralph Kimball bottom-up Data Warehouse Model
In Ralph Kimball’s methodology, the bottom-up process is the result of an initial study. Focused on an analysis of the relevant business processes to be modelled.
Ralph Kimball is famous for his excellent book: The Microsoft Data Warehouse Toolkit.
And as being the founder of the Kimball Group.
Data Warehouse modelling with specialised data shops
Firstly, the data shops contain the dimensions of the analysis axes and the facts or measures. Facts contain either atomic data, i.e., at a fine level, or aggregated data. It often models a very specific area of activity such as sales or production.
These data shops can be integrated with a business intelligence solution to create a complete data warehouse.
This integration between different data warehouses is implemented by what Kimball calls a data warehouse bus architecture. Only a collection of compliant dimensions allows this integration.
These are dimensions that are shared (in a specific way) between two or more data repositories. They allow cross-analysis across multiple business or operational domains.
Ralph Kimball’s bottom-up modelling enables Drill Across
Secondly, conforming dimensions representing entry points between data marts enable cross-data integration.
The actual integration of two or more data marts is then achieved through a process called “Drill Across”. This is a lateral drilling that groups different functional data but at the same level of granularity. It therefore uses the same dimensions.
The dimensions often used because they are transversal to the company are, for example, the axes:
Maintaining accurate management of the data warehouse architecture is essential for data integrity. The most important management task is to ensure that the dimensions between data marts are compatible. To enable parallel updates.
Some analysts believe that segmentation of the data warehouse is an advantage of the Kimball method. This is done in a few logical and coherent data marts, rather than in a large, centralised and often complex model.
A connected silo view that supports BI project iterations
Thirdly, another advantage is the analysis of the company’s data as soon as the first data shops have been created. The method also allows an exploratory and iterative approach to building the warehouse.
For example, capitalising on the effort to develop data loading for sales, with a specific data warehouse such as that for production and stocks.
The rest of the BI project can continue for Production, a joint analysis can highlight the correlation between production capacity and daily sales.
This Ralph Kimball view of the BI world is therefore very conducive to iterations and allows the production of BI projects domain by domain.
Find more on Business Intelligence projects with this article on Decision Support System definition.