ALECE An Attention-based Learned Cardinality Estimator for SPJ Queries on Dynamic Workloads (Extended)

Query-driven and data-driven methods

• query-driven: it basically trains a model to predict CE.
• data-driven: it basically learns a join distribution among columns and then sampling based on that, and use the sampled data to estimate the basic statistics. The default optimizer will then use those statistics to estimate CE.

Existing work:

• Cannot combine both query-driven and data-driven methods.
• Cannot handle dynamic workloads that mix queries and data manipulation statements including inserts, deletes, and updates.

ALICE is less sensitive to data changes

---

• ← Previous Post
• Next Post →