Reproduce “AI-Driven Research for Databases” 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
Audrey Cheng, Harald Ng, Aaron Kabcenell, Peter Bailis, Matei Zaharia, Lin Ma, Xiao Shi, Ion Stoica
A co-evolutionary framework employs Large Language Models to autonomously refine both database optimization algorithms and their evaluation pipelines, addressing a critical bottleneck in AI-Driven Research for Systems. This method discovered new buffer cache, index selection, and query rewrite policies, demonstrating performance gains up to 19.8% higher buffer cache hit rate and 4.0x faster query execution for DSB queries over state-of-the-art baselines.
Key insights
- "More is more": Maximizing simulator fidelity by calibrating against broader ground-truth baselines and exposing rich system state guides AI to discover complex, high-performing algorithms.
- "Mind the gap": Establishing a reliable fitness signal requires actively investigating and bridging discrepancies between proxy metrics (e.g., optimizer cost estimates) and true end-to-end performance (e.g., query latency).
- "Go off what you know": Efficiently generating deployable policies involves exploiting prior empirical successes to guide workload selection and search space pruning, progressively expanding into beneficial areas.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.