Reproduce “Training-Free Query Optimization via LLM-Based Plan Similarity” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nikita Vasilenko, Alexander Demin, Vladimir Boorlakov
The LLM-PM framework introduces a training-free method for automatic query hint selection in database systems by leveraging pre-trained Large Language Model embeddings of execution plans. This approach achieved an average 21.1% total aggregate speed-up in query latency on the JOB-CEB workload with OpenGauss, primarily by optimizing the slowest queries.
Key insights
- LLM embeddings of execution plans provide a more semantically rich and effective representation for plan similarity than raw SQL query text or simple string-based plan comparisons.
- A training-free, retrieval-based approach using pre-trained LLM embeddings can deliver substantial performance improvements without the typical computational overhead and latency of complex ML models.
- The two-neighborhood consistency check is vital for the practical application of LLM-based hint selection, significantly improving performance and preventing detrimental hint suggestions by acting as a safety filter.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.