ScholarWorks@숙명여자대학교

초록

SQL query optimization, essential for improving database system performance, remains an active area of research. Traditional cost-based optimization methods, however, are limited in accurately predicting actual execution costs due to statistical errors and imprecise cardinality estimations, especially with increasing data scales and query complexity. This study proposes a Hybrid-GM model that integrates the graph structure of graph neural network (GNNs) with global query feature statistics. This model employs an multi-layer perceptron (MLP) encoder to capture global query features and a GNN encoder to learn the structural properties of query execution plans, enabling accurate prediction of query execution costs. The Hybrid-GM model demonstrated improved performance compared to standalone MLP and GNN models, achieving relative reductions of 74.3% and 68.6% in root mean squared error (RMSE) and mean absolute error (MAE), respectively. This study is expected to make a significant contribution to the advancement of machine learning-based query optimization.

키워드