Lero A Learning-to-Rank Query Optimizer

Latency prediction suffers from unstable performance, high training cost, and slow model updating.

Solutions

Relative order or rank of execution plans, rather than the exact cost or latency is sufficient for query optimization.

• The classification task is much easier than the regression tasks.
• Non-intrusive design.

Some Details

cardinality spans in wide range, it uses min-max normalization.

plan embedding model and comparator model are learned together.

Plan enumerator: it is tuned by changing cardinalities.

###

---

• ← Previous Post
• Next Post →