| Data warehouses collect data from one or more external sources and translate it to a common schema that is easily queryable. In contrast to traditional on-line transaction processing (OLTP) database systems in which clients perform a mix of short-duration read and update transactions on the database, warehouse clients typically perform complex read-only queries, in order to analyze the data for trends and anomalies. This type of query processing is often referred to as on-line analytical processing (OLAP). In order to speed up the evaluation of such complex queries, warehouses usually precompute and store the results of certain queries. The stored results are called materialized views, and often involve aggregating data from large base relations.; As changes are made to the source base relations, the warehouse views must be updated. Source changes are often applied to the warehouse views at regular intervals, usually once a day, in a large batch. Maintaining views in a data warehousing environment incurs problems not normally encountered when views are stored in the same database as the base data: First, extra information not available in the view itself is often required to maintain the view. This information must be obtained by querying the sources or by storing additional information in the warehouse. Second, aggregation is especially common in warehouse views; therefore, efficient algorithms for maintaining views are critical in data warehousing environments. Third, while the views are being maintained, the warehouse is often made unavailable to readers, which is unacceptable for global warehouses that need to be accessed around the clock. Algorithms are needed that reduce or eliminate the amount of time the warehouse is unavailable during view maintenance. This thesis makes important contributions in each of the above areas. |
