This is Going to Sound Crazy, But What If We Used Large Language Models to Boost Automatic Database Tuning Algorithms By Leveraging Prior History? We Will Find Better Configurations More Quickly Than Retraining From Scratch!
Summary: Booster augments existing DBMS autotuners with LLMs and query-level historical artifacts, turning prior tuning runs into per-query config contexts that can be reused under workload drift/schema transfer. Beam-search composition of query suggestions yields much faster re-optimization, up to 74% better configs than retraining/continuing from scratch.
(summarized by gpt-5-mini on Apr 11 2026)
@inproceedings{zhang_sigmod26,
title = {{This is Going to Sound Crazy, But What If We Used Large Language Models to Boost Automatic Database Tuning Algorithms By Leveraging Prior History? We Will Find Better Configurations More Quickly Than Retraining From Scratch!}},
author = {Zhang, William and Lim, Wan Shen and Pavlo, Andrew},
series = {{SIGMOD} '26},
booktitle = {Proceedings of the {ACM} {SIGMOD} International Conference on Management of Data},
publisher = {Association for Computing Machinery},
doi = {10.1145/3786704},
url = {https://dl.acm.org/doi/10.1145/3786704},
year = {2026}
}