OLAP Concepts

There are three OLAP components. They are the data mart, the data warehouse and the enterprise data warehouse. Data marts support a business unit such as a department. Data warehouses support large or complex business units such as a division. Enterprise data warehouses support the corporation or government agency.

All three OLAP components contain dimensions and fact tables organized in a multi-dimensional data architecture. Building the OLAP component is an "evolutionary process", beginning with a single data mart and growing it into a data warehouse, one step at a time.

The simplest OLAP implementation is the data mart. The data mart contains a single fact table and two or more dimensions. The data mart supports a single business intelligence application or business model. A single data mart should be the first OLAP structure built.

Next is the data warehouse. The data warehouse contains many fact tables and many dimensions. The data warehouse supports many business intelligence applications or business models. The data warehouse exists once a second fact table is added to the original data mart. The data warehouse grows as new dimensions and fact tables are incrementally added.

The largest and most ambitious OLAP implementation is the enterprise data warehouse. The enterprise data warehouse contains all of the fact tables and dimensions known to the corporation or government agency. The enterprise data warehouse supports all of the business intelligence applications and models of the corporation or agency. The enterprise data warehouse is logical conclusion of incrementally adding dimensions and fact tables to a data warehouse or merging two or more data warehouses.

A dimension is a denormalized table created from many OLTP components. A dimension violates the 3rd Normal Form, typically containing many non-functional dependencies. A dimension contains mostly descriptive attributes from the Type, Class, Entity and Relation OLTP components and may also contain aggregate attributes from associated fact tables. There is one dimension for each Entity in the OLTP data models or Class Matrix. There is no justification for having more than one Customer or Product dimension in an enterprise. As a matter of fact, having more than one Customer or Product dimension may create conflict and confusion by producing conflicting business intelligence models.

A fact table is a normalized table containing additive data from OLTP Instance, Detail and Journal components. Additive data is used for calculations, such as QuantityOrdered and QuantityShipped. Semi-additive data, such as LineItemTotal, and non-additive data, such as ProductDescription, should never be used as fact table attributes. Additive data is also atomic, which ensures that the fact table is built at its "lowest level of granularity." A data mart is defined by a fact table.