"What makes my queries slow?": Subgroup Discovery for SQL Workload Analysis

Automated pattern discovery can help database administrators identify performance problems more efficiently: Rather than manually analyzing thousands of queries, the researchers developed a technique to automatically discover which combinations of query characteristics (like specific SQL clauses or system metrics) are associated with slow performance, potentially saving DBAs significant time in troubleshooting.

Explainability is critical for AI tools used in technical operations: The paper emphasizes that discovered patterns should be presented as "hypotheses" rather than black-box recommendations, allowing database administrators to understand why queries are slow and verify findings themselves—a key principle of responsible AI in high-stakes technical environments.

Real-world testing on production data validates practical utility: The researchers tested their approach on actual workload data from hundreds of company databases over a full week, demonstrating that the technique can discover "insightful hypotheses" in real operational settings rather than just in controlled experiments.

Transparency through visualization supports human decision-making: The team built an interactive visualization tool alongside their analysis method, recognizing that DBAs need to explore and understand discovered patterns themselves rather than simply receiving automated recommendations, promoting human oversight of AI-assisted decisions.

Making datasets and code publicly available enables scrutiny and reproducibility: By releasing both their dataset and source code, the researchers support the responsible AI principle of transparency, allowing others to verify their findings and build upon their work.