Refinamento das estimativas de cardinalidade no processamento de consultas

Abstract

Database Management Systems (DBMSs) make use of a declarative language in order to allow queries to stored data to be performed. In this approach, users describe through a textual query what they want to get from DBMSs, leaving them to define how data will be processed and ultimately retrieved. Therefore, DBMSs must choose from the different possibilities of executing a given query which one, based on an estimation process, is the one with the best performance. Cardinality is the number of tuples that result from applying a relational operator to a relation. This information is estimated by the DBMSs to calculate the number of tuples that result from one operation and serve as input to another in the processing of a query. The optimization process, which seeks to generate the best way to process a query, makes use of these estimated cardinalities to make optimization decisions, such as choosing predicate ordering. That said, through this dissertation, we propose a new approach to calculate the cardinality estimates of query operations in order to guide the execution engine of the DBMSs to make the correct choice of the way the query will be executed, that is, choose the best possible form or at least avoid the worst one. We believe that adding the use of Machine Learning models in the process of estimating query cardinality should lead to more efficient query execution. To ensure this hypothesis, experimental tests were performed using the PostgreSQL. Finally, it is possible to conclude from the preliminary experiments presented in this work that this new approach may result in improvements in query processing in DBMSs, especially in the generation of cardinality estimates.