Reproduce “E2ETune: End-to-End Knob Tuning via Fine-tuned Generative Language Model” 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
Xinmei Huang, Haoyang Li, Jing Zhang, Xinxin Zhao, Zhiming Yao, Yiyan Li, Tieying Zhang, Jianjun Chen and 2 more
E2ETune introduces an end-to-end database knob tuning method utilizing a fine-tuned generative language model to directly map workload characteristics to optimal configurations. This approach reduces tuning time by up to 98.6% compared to iterative methods, delivering competitive or superior performance across diverse benchmarks and real-world workloads.
Key insights
- Generative language models possess the capability to learn complex, implicit mappings between workload characteristics and optimal database configurations, enabling effective end-to-end tuning without iteration.
- Employing a cost model to simulate performance during offline data collection significantly reduces the time and resources needed to generate high-quality training labels for data-hungry language models.
- Comprehensive workload features (statistics, query plans, internal metrics) and a discretized, bucketed output format for knob values are critical for a generative LM to achieve strong performance and generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.