Query-condition-aware V-optimal histogram in range query selectivity estimation

Obtaining the optimal query execution plan requires a selectivity estimation. The selectivity value allows to predict the size of a query result. This lets choose the best method of query execution. There are many selectivity obtaining methods that are based on different types of estimators of attribute values distribution (commonly they are based on histograms). The adaptive method, proposed in this paper, uses either attribute values distribution or range query condition boundaries one. The new type of histogram - the Query-Conditional-Aware V-optimal one (QCA-V-optimal) - is proposed as a non-parametric estimator of a probability density function of attribute values distribution. This histogram also takes into account information about already processed queries. This information is represented by the 1-dimensional Query Condition Distribution histogram (HQCD) which is an estimator of the include function PI which is also introduced in this paper. PI describes so-called regions of user interest, i.e. it shows how often regions of attribute values domain were used by processed queries. Advantages of the proposed method based on QCA-V-optimal are presented. Conducted experiments reveal small values of a mean relative selectivity estimation error comparing to the error values obtained by methods based on the relevant classical V-optimal histogram and Equi-height one.