Incremental maintenance of materialized views

Abstract

Materialized views are important in databases, particularly in data warehouses, where they are used to speed up query processing on large amounts of data. These views need to be maintained incrementally in response to database updates. In this paper we investigate the problem of incremental maintenance of a materialized view in response to changes to the base data. We show that a materialized view can be maintained without accessing the view itself by materializing additional relations. We firstly give an algorithm for determining what additional relations need to be materialize in order to maintain a view incrementally. We then propose an algorithm for updating the materialized view (and the additional relations) based on the optimized operator tree used for evaluating the view as a query. A materialized view is updated by propagating updates to the nodes of its operator tree in a bottom-up fashion, where each node contains a relational algebraic operator and computes an intermediate expression. Our algorithm derives the exact update at every intermediate node in the tree, including the materialized view. Finally, we compare our incremental algorithm with the naive algorithm that recomputes the view from scratch.

Part of work done while visiting Kyoto University and supported by JSPS.