Scholarlink

<p>With the rise of artificial intelligence in database query optimization, learning-based methods have become a promising approach to improving query performance. However, current learned optimizers struggle to model the complex structures of candidate plans and adapt to dynamic workloads. To address these issues, we propose Saro+, an enhanced learning-based optimizer building on our prior work, Saro. Saro+ introduces Tree Node Dependency Features to better capture execution plan structure and a Hybrid Ranking Model that combines listwise and pairwise strategies, balancing global and local optimization. Experiments on STATS and IMDb show that Saro+ outperforms both the PostgreSQL optimizer and single-strategy learned methods in query latency and stability, demonstrating strong robustness and adaptability under high concurrency and workload variability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing query optimization with hybrid ranking model and tree node dependency

  • 期刊论文

  • 发表日期：

    2025年9月25日

查看全文
  • Chong Sun,
  • Shipeng Cao,
  • Jiangqing Wang,
  • Lu Zheng,
  • Jin Xie,
  • Chongwei Ruan,
  • Haijiao Wang,
  • Ziheng Han

摘要

With the rise of artificial intelligence in database query optimization, learning-based methods have become a promising approach to improving query performance. However, current learned optimizers struggle to model the complex structures of candidate plans and adapt to dynamic workloads. To address these issues, we propose Saro+, an enhanced learning-based optimizer building on our prior work, Saro. Saro+ introduces Tree Node Dependency Features to better capture execution plan structure and a Hybrid Ranking Model that combines listwise and pairwise strategies, balancing global and local optimization. Experiments on STATS and IMDb show that Saro+ outperforms both the PostgreSQL optimizer and single-strategy learned methods in query latency and stability, demonstrating strong robustness and adaptability under high concurrency and workload variability.