Query optimization is a cornerstone of database management systems, critical for efficient data retrieval and processing. This project explores the capabilities and limitations of Bao, a learned query optimizer for PostgreSQL, by testing it on high-complexity benchmarks. We aim to assess Bao’s adaptability and performance in novel scenarios such as skewed data, skewed queries, and dynamic workloads. Our experiments target optimizations in disk I/O, energy efficiency, and cache management while analyzing Bao’s robustness against changes in data and query characteristics.
- Modified ImDb queries**
IC: Vidhi Rambhia- Analyze how Bao handles schema evolution and dynamic changes (slightly modified queries).
- Visualize results through graphs showcasing the impact on performance.
- Skew Data
IC: Saharsh Barve- Test bao on queries with mixed selectivity workloads (high selectivity | low selectivity).
- Train Bao on one cluster and evaluate its accuracy on others.
- Study performance metrics like execution efficiency and query accuracy.
- Schema Changes
- Dropped indexes to evaluate resilience to structural modifications
- Investigate how this impacts optimization strategies and execution plans.
IC: Vidhi Rambhia
IC: Rahul Bothra
- Focus on:
- Strategies that have "broken" query optimizers.
- Comparing our work with Bao and its benchmarks without redundancy.
IC: Omkar Dhekane
- Prepare Slides for project presentations.
- Vidhi Rambhia: vidhisr2@illinois.edu
- Rahul Bothra: bothra2@illinois.edu
- Saharsh Barve: ssbarve2@illinois.edu
- Omkar Dhekane: dhekane2@illinois.edu
- Bao Tutorial: https://rmarcus.info/bao_docs/tutorial/2_bao_setup.html
- Bao paper: https://arxiv.org/pdf/2004.03814
- Bao github repo: https://github.com/learnedsystems/BaoForPostgreSQL