Access Plan Recommendation Using SQL Queries Similarity

摘要

Plan reuse is a technique of databases optimization, the main purpose of which is to reuse old access plans stored in the database to execute future queries instead of generating new plans. To carry out its task, the optimizer needs to identify similarity between new and old queries. Questions such as which techniques are needed and which SQL query representation is best to produce accurate similarity estimation remain poorly addressed. The main goals of this work is to propose an approach for access plan recommendation using 4 SQL queries representations and clustering techniques to identify similarity between queries. We study SQL queries similarity, at the intentional level by considering the uninterpreted SQL sentences; therefore SQL queries are represented to TF-IDF and N-GRAM. Then, feature selection algorithms are used to identify the most significant descriptors in the feature space. Next, clustering is applied using partitionning clustering, density based clustering and competitive learning based clustering. Finally, a access plan is recommended to the optimizer. Results show that the use of feature selection process is relevant to our work especially for TF-IDF representation while the most accurate and efficient clutsering is obtained with k-means algorithm.