This is an old revision of the document!
Relational and object-oriented databases are mainly suited for operational settings in which there are many small transactions querying and writing to the database. Consistency of the database (in the presence of potentially conflicting transactions) is of utmost importance. Much different is the situation in analytical processing where historical data is analyzed and aggregated in many different ways. Such queries differ significantly from the typical transactional queries in the relational model:
For these reasons, data to be analyzed is typically collected into a data warehouse with Online Analytical Processing support. Online here refers to the fact that the answers to the queries should not take too long to be computed. Collecting the data is often referred to as Extract-Transform-Load (ELT). The data in the data warehouse needs to be organized in a way to enable the analytical queries to be executed efficiently. For the relational model star and snowflake schemes are popular designs. Next to OLAP on top of a relational database (ROLAP), also native OLAP solutions based on multidimensional structures (MOLAP) exist. In order to further improve query answering efficiency, some query results can already be materialized in the database, and new indexing techniques have been developped.
In the course, the main concepts of multidimensional databases will be covered and illustrated using the SQL Server tools. Complimentary to the course, IBM and Teradata will give invited lectures.
The following materials have been used to construct the course material, but are not required reading for the course: