Query optimizer is one of the key components which determines DBMS performance under OLAP workload. Nevertheless, it was shown that query optimizer often fails to find a good execution plan because of incorrect cardinality estimations.
The perspective approach to improve cardinality estimation quality is adaptive query optimization. In contrast with classical approaches, which rely on the precomputed histograms, it utilizes the execution statistics of the previously executed queries to refine cardinality estimations. However, the original AQO uses a modification of kNN machine learning method, which implies a number of limitations on the AQO applicability.
In the current lecture we will make an introduction to the kNN-based AQO. Afterwards, we will present you our novel neural network-based AQO which can potentially overcome the limitations of the kNN-based AQO, and will demonstrate the first experimental results.