Performance evaluation of incremental materialized view maintenance in ORDBMS

Abstract

A materialized view is a derived relation stored in the database, resembles like tables and behaves like indexes. Because of the query intensive nature of data warehousing or online analytical processing applications, materialized view is quite promising in efficiently processing queries to improve query performance. When a base relation is updated, all its dependent materialized views have to be updated in order to maintain the consistency and integrity of the database in response to the changes in the base relation. It is costly to rematerialize the view each time a change is made to the base tables that might affect it and it is desirable to propagate the changes incrementally. Hence, all of the views cannot be materialized due to the maintenance cost. So, it is necessary to evaluate the performance of incremental materialized view maintenance and to determine the circumstances in which a view is beneficial to be materialized for faster query performance. It is also necessary to dynamically select a subset of views from a set of views queried at a particular time period based on the query processing cost and view maintenance cost. A methodology has been developed based on the performance affecting factors like - view selectivity, complexity and database size to evaluate the performance of incremental view maintenance and to determine the situations a view is profitable for materialization by computing the incremental propagation cost, query answering cost and relative costs of query answering versus propagating a materialized view. After this a dynamic cost model has been designed incorporating the above mentioned factors as well as query access frequency, execution time, table update frequency and view maintenance cost to select a subset of views from a set of views for materialization and to replace the old materialized views that are no longer in use or the materialized view access frequency is too low. A number of algorithms have been designed and mathematical equations have been developed to define the dynamic threshold level. At the end, experimental results have been carried out for the incremental maintenance performance evaluation and on dynamic view selection and removal by using synthetic and real data sets with different characteristics in object-relational database. The outcome of the thesis reveals that the incremental maintenance is always cost effective. Finally, dynamic view selection for materialization and removal of old materialized views is explored based on dynamic threshold level.