BASE: Bridging the Gap between Cost and Latency for Query Optimization

• DOI: 10.14778/3594512.3594525
• Publication Date: 2023-06-23T00:28:36Z
• Authors: Xu Chen; Zhen Wang; Shuncheng Liu; Yaliang Li; Kai Zeng; Bolin Ding; Jingren Zhou; Han Su; Kai Zheng
• Keywords: Benchmark (surveying)

Some recent works have shown the advantages of reinforcement learning (RL) based learned query optimizers. These often use cost (i.e., estimation model) or latency execution time) as guidance signals for training their models. However, cost-based underperforms in and latency-based is time-intensive. In order to bypass such a dilemma, researchers attempt transfer value network from domain domain. We recognize critical insights cost/latency-based training, prompting us reward function rather than network. Based on this idea, we propose two-stage RL-based framework, BASE , bridge gap between latency. After policy its first stage, formulates transferring variant inverse learning. Intuitively, learns calibrate updates regarding calibrated one mutually-improved manner. Extensive experiments exhibit superiority two benchmark datasets: Our optimizer outperforms traditional DBMS, using 30% less time SOTA methods. Meanwhile, our approach can enhance efficiency other learning-based