A data mart is generally a relatively small application- or function-specific subset of the data warehouse database created to optimize application performance for a narrowly defined user population.
Data marts are often categorized into three different types:
Independent data marts are isolated entities, entirely separate from the enterprise data warehouse. Their data derives from independent sources and they should be viewed as data pirates in the context of the enterprise data warehouse because their independent inputs, which are entirely separate from the enterprise data warehouse, have a high likelihood of producing data that does not match that of the warehouse.
These independent data marts are sometimes referred to as data basements, and Teradata strongly discourages their use (see “Independent Data Marts” on page 20).
Dependent data marts are derived from the enterprise data warehouse. Depending on how a dependent data mart is configured, it might or might not be useful.
The recommended process uses only data that is derived from the enterprise data warehouse data store and also permits its users to have full access to the enterprise data store when the need to investigate more enterprise-wide issues arises.
The less useful forms of dependent data mart are sometimes referred to as data junkyards (see “Dependent Data Marts” on page 22).
The logical mart is a form of dependent data mart that is constructed virtually from the physical data warehouse. Data is presented to users of the mart using a series of SQL views that make it appear that a physical data mart underlies the data available for analysis (see “Logical Data Marts” on page 23).