17/04/2024

Tech Update

The Best Tech Research

Components of a Data Warehouse Architecture – Part 2, The Kimball Presentation Area

Components of a Data Warehouse Architecture – Part 2, The Kimball Presentation Area

In part 1 of this article series, we described the staging area and the ETL process of a data warehouse architecture. In the present and following article we shall describe the presentation area of the data warehouse. The term presentation is used to denote the fact that this is the area, where data are presented to its Customers (the business analysts). There is no globally acceptable standard on the development of the data warehouse presentation area. Two major approaches have prevailed:· the dimensional datawarehouse approach (proposed by R. Kimball)· the corporate information factory (CIF) approach (proposed by B. Inmon) Kimball approach According to the Kimball approach, the presentation area is made of a number of dimensional data structures, called star schemas. A star schema is a relational data structure which involves the following:

  • a fact table which stores all the measurements used to produce performance analytics and is linked to
  • a set of tables which capture the dimensional information, related to the above mentioned measurements.

A set of linked star schemas which focus on a single business process, are called a data mart. Star schemas are linked to each other, based on conformed dimension tables (according to the Kimball parlor, this is called a bus architecture). The major advantage of the dimensional data structure is derived from the symmetry and simplicity of the star schema, which:

  • is easily understood by business users, who can access and use it directly without database manipulation skills
  • performs better in the execution of complex queries, than complex normalized data structures (used commonly in operational systems)

The Kimball approach is criticized on the following points:

  • how feasible it is to connect data marts which capture information at a different level of detail. Linking of data marts is very important, since it allows the combined analysis of data from different marts (a practice known as drill across).
  • normalized database structures reflect better complex entity type relationships, compared to the denormalised dimensional model. The Kimball approach proposes the exclusive use of denormalised structures.

Copyright 2006 – Kostis Panayotakis