A Self-Maintainable View Maintenance Technique for Data Warehouses

Materialised views are important in data warehouses where they are used to speed up query processing in decision support activities. There are many applications, such as those involving financial or stock market data for example, where the base data is volatile and the updates to the materialised views need to be computed efficiently. In such cases, incremental update techniques are often used for performance reasons. In this paper we propose an incremental technique for efficiently maintaining materialised views in these high performance applications by materialising additional relations which are derived from the intermediate results of the view computation. We present an algorithm which uses the operator tree for the view to determine which additional relations need to be materialised in order to maintain the view. We then give an incremental algorithm for maintaining both the view and the additional relations which has several desirable features. Firstly, it is an 'exact change' algorithm. By this we mean that if the deletion of a tuple is produced by our technique then the tuple is guaranteed not to exist in the new view, and likewise, any tuple to be inserted into the view is guaranteed not to exist in the old view. Secondly, the view and the additional relations are self-maintainable. Lastly, our technique allows multiple updates to the base relations and computes the effects of these updates simultaneously to the view.