A different approach to the index selection problem is proposed, with partial indexes as the focal point. FP-Growth, as well as a dynamic programming algorithm for optimization, is adopted from a previous research project, which also serves as a basis for the proposed solution. The aim of this research is to minimize wastage of memory and, ultimately, allow for more indexes to expedite the processing of incoming queries. Instead of suggesting potential indexes based solely on which attributes appear most frequently in the workload of queries, the density of filtered queries are also accounted for, i.e. columns of corresponding attributes are only indexed in smaller portions as long as these portions satisfies the majority of all queries for said attributes. MAFIA, a subspace clustering technique based on CLIQUE, has been implemented in order to discover these dense areas. To evaluate the effectiveness of the solution, a comparison to the former solution was made through numerous experiments, and the performance brought about by the generated indexes was benchmarked and compared between different test-cases.
抄録全体を表示