Selection and Pruning Algorithms for Bitmap Index Selection Problem Using Data Mining

Abstract

Indexing schemes are redundant structures offered by DBMSs to speed up complex queries. Two types of indices are available: mono-attribute indices (B-tree, bitmap, hash, etc.) and multi-attribute indices (join indices, bitmap join indices). In relational data warehouses, bitmap join indices (BJIs) are bitmap indices for optimizing star join queries through bit-wise operations. They can be used to avoid actual joins of tables, or to greatly reduce the volume of data that must be joined, by executing restrictions in advance. BJIs are defined using non-key dimension attributes and fact key attributes. Moreover, the problem of selecting these indices is difficult because there is a large number of candidate attributes (defined on dimension tables) that could participate in building these indices. To reduce this complexity, we propose an approach which first prunes the search space of this problem using data mining techniques, and then based on the new search space, it uses a greedy algorithm to select BJIs that minimize the cost of executing a set of queries and satisfy a storage constraint. The main peculiarity of our pruning approach, compared to the existing ones (that use only appearance frequencies of indexable attributes appearing in queries as a pruning metric), is that it uses others parameters such as the size of their dimension tables, the length of each tuple and the size of a disk page.