An extensible model of selectivity estimation

Abstract

We present an extensible model of selectivity estimation for the relational algebra select operator. The model permits a flexible handling of distribution and correlation assumptions, relational expression types, and abstract data types. The main feature of the model is the separation of selectivity methods from the estimation process. This allows the database designer to select the methods which best fit the requirements of a particular database. We present in detail the process of defining estimation methods and associating them with attributes and the general procedure of estimating tuple cardinality. Estimating tuple cardinality involves the analysis of a Boolean expression and the selection of applicable estimation methods. We also discuss two approaches for applying an extensible model to a tree of relational algebra operations and show how our select model fits into these larger frameworks.