Reproduce “A Novel Technique for Query Plan Representation Based on Graph Neural Nets” 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
Baoming Chang, Amin Kamali, Verena Kantere
Researchers at the University of Ottawa developed BiGG, a novel Graph Neural Network-based model for query plan representation, demonstrating its superior ability to capture complex plan intricacies. BiGG significantly enhanced cost estimation accuracy, achieving a Median Q-error of 1.762 and a Spearman's Correlation of 0.805 on complex TPC-DS workloads, surpassing prior state-of-the-art methods.
Key insights
- The choice of query plan representation model significantly influences cost estimation accuracy, particularly under complex workloads, challenging previous claims of negligible impact.
- Bidirectional Graph Neural Networks with weighted directed edges are crucial for capturing intricate structural and logical dependencies in query plans, leading to more accurate representations.
- Aggregating GNN-learned node features using a GRU in post-order traversal effectively preserves structural and valuable execution sequence information, improving overall representation quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.