A Data-Aware and Adaptability-Enhanced Framework for Cardinality Estimation

To achieve efficient query processing, DBMS query optimizers have long relied on precise cardinality estimation methods. To enhance the accuracy of cardinality estimation, we introduce DACE, an Cardinality Estimation framework based on Data-Aware and Adaptability-Enhanced. First, we propose a data-aware dual attention cardinality estimator that uncovers implicit relationships between queries and underlying data based on the characteristics of the queries and data. Subsequently, we develop a learning strategy utilizing Bayesian neural networks and active learning to build an ensemble model that enhances the adaptability of the estimation model through various data sampling strategies and improves end-to-end execution efficiency. Lastly, experiments conducted on multiple workloads demonstrate the effectiveness and efficiency of the proposed framework.