A Bayesian Approach to Estimating the Selectivity of Conjunctive Predicates

Cost-based optimizers in relational databases make use of data statistics to estimate intermediate result cardinalities. Those cardinalities are needed to estimate access plan costs in order to choose the cheapest plan for executing a query. Since statistics are usually collected on single attributes only, the optimizer can not directly estimate result cardinalities of conjunctive predicates over multiple attributes. To avoid having to fall back to assuming statistical independence, modern relational database systems offer the possibility to additionally collect joint statistics over multiple attributes. These statistics allow a direct cardinality estimate for conjunctive predicates. A widely used approach is collecting the number of distinct value combinations as a joint statistic. This can be used for a uniformity based estimate, which assumes each value combination to occur equally often. Although this estimate is likely an improvement, it is still inaccurate, since “real world” data is unlikely to be uniform. This paper proposes a new approach of estimating the result cardinality of conjunctive predicates over multiple attributes of a relation. The proposed method combines knowledge from single-column histograms using a conditional probability based “uniform correlation” approach. Initial evaluation shows that this method yields better results for estimating predicates on highly correlated attributes than the classic uniformity based approach.