Efficient Incremental View Maintenance for Data Warehousing

Data warehousing and on-line analytical processing (OLAP) are essential elements for decision support applications. Since most OLAP queries are complex and are often executed over huge volumes of data, the solution in practice is to employ materialized views to improve query performance. One important issue for utilizing materialized views is to maintain the view consistency upon source changes. However, most prior work focused on simple SQL views with distributive aggregate functions, such as SUM and COUNT. This dissertation proposes to consider broader types of views than previous work. First, we study views with complex aggregate functions such as variance and regression. Such statistical functions are of great importance in practice. We propose a workarea function model and design a generic framework to tackle incremental view maintenance and answering queries using views for such functions. We have implemented this approach in a prototype system of IBM DB2. An extensive performance study shows significant performance gains by our techniques. Second, we consider materialized views with PIVOT and UNPIVOT operators. Such operators are widely used for OLAP applications and for querying sparse datasets. We demonstrate that the efficient maintenance of views with PIVOT and UNPIVOT operators requires more generalized operators, called GPIVOT and GUNPIVOT. We formally define and prove the query rewriting rules and propagation rules for such operators. We also design a novel view maintenance framework for applying these rules to obtain an efficient maintenance plan. Extensive performance evaluations reveal the effectiveness of our techniques. Third, materialized views are often integrated from multiple data sources. Due to source autonomicity and dynamicity, concurrency may occur during view maintenance. We propose a generic concurrency control framework to solve such maintenance anomalies. This solution extends previous work in that it solves the anomalies under both source data and schema changes and thus achieves full source autonomicity. We have implemented this technique in a data warehouse prototype developed at WPI. The extensive performance study shows that our techniques put little extra overhead on existing concurrent data update processing techniques while allowing for this new functionality