ALECE An Attention-based Learned Cardinality Estimator for SPJ Queries on Dynamic Workloads (Extended)
1 minute read ∼ Filed in : A paper noteQuery-driven and data-driven methods
- query-driven: it basically train a model to predict CE.
- data-driven: it basically learn a join distribution among columns and then sampling bsaed on that, and use the sampled data to estimate the basic staisitcs. 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.
ALECE is less sensitive to data changes
Techniques