Improving quality and convergence of genetic query optimizers

The application of genetic programming strategies to query optimization has been proposed as a feasible way to solve the large join query problem. However, previous literature shows that the potentiality of evolutionary strategies has not been completely exploited in terms of convergence and quality of the returned query execution plans (QEP). In this paper, we propose two alternatives to improve the performance of a genetic optimizer and the quality of the resulting QEPs. First, we present a new method called Weighted Election that proposes a criterion to choose the QEPs to be crossed and mutated during the optimization time. Second, we show that the use of heuristics in order to create the initial population benefits the speed of convergence and the quality of the results. Moreover, we show that the combination of both proposals outperforms previous randomized algorithms, in the best cases, by several orders of magnitude for very large join queries.