Online Analytical Processing

Data marts and data warehouses support OLAP systems. Online analytical processing systems (OLAP) produce business intelligence in the form of reports, spreadsheets and visual business models.

The difference between OLTP and OLAP is similar to the difference between a photograph and a movie. A photograph provides great detail about a moment in time, while a movie provides great detail over a period of time.

The goal of OLTP is to provide everything that is known about a particular event. The goal of OLAP is to provide everything that is known about many events, usually in relation to time.

This requires different data architectures and processing. OLTP may insert or update a few table rows to achieve its goal, while OLAP may query millions of table rows to achieve its goal.

To distinguish between OLTP and OLAP, the industry has adopted the terms "database, data mart" and "data warehouse". OLTP is supported by databases using a relational architecture. OLAP is supported by data marts and data warehouses using a multi-dimensional architecture. However, as far as the database management system is concerned, they are all databases.

OLTP databases support business operations and their designs must be dynamic in order to support high volume, event-based application systems which are transaction-oriented and change often due to new technology and business or marketplace requirements. As a result, OLTP systems must be supported by databases that are easy to change and structured for high volume, random insert and update. The relational database architecture is highly suitable for this purpose.

OLAP systems which are supported by data marts and data warehouses are more stable and are structured for high volume query. The multi-dimensional architecture is ideal for querying gigabytes or terabytes of data.

Another important difference between the architectures is that OLTP data are automatically inserted and updated by the associated applications system while OLAP data need to be extracted from the OLTP databases, cleansed of processing errors and exceptions, reformatted and loaded into the data warehouse.