A heuristic approach for optimization of path expressions

Abstract

The object-oriented database management systems store references to objects (implicit joins, precomputed joins), and use path expressions in query languages. One way of executing path expressions is pointer chasing of precomputed joins. However it has been previously shown that converting implicit joins to explicit joins during the optimization phase may yield better execution plans. A path expression is a linear query, therefore, considering all possible join sequences within a path expression is polynomial in the number of classes involved. Yet, when the implicit joins are converted to explicit joins in a query involving multiple path expressions bound to the same bind variable, the query becomes a star query and thus considering all possible joins is exponential in the number of paths involved. This implies that there is a need for improvement by using heuristic in optimizing queries involving multiple path expressions.

A heuristic based approach for optimizing queries involving multiple path expressions is described in this paper. First, given the cost and the selectivities of path expressions by considering a path expression as a unit of processing, we provide an algorithm that gives the optimum execution order of multiple path expressions bound to the same bind variable. For this purpose, we derive the formulas for the selectivities of path expressions. Then by using this ordering as a basis we provide a general heuristic approach for optimizing queries involving multiple path expressions.

Two optimizers are developed to compare the performance of the heuristic based approach suggested in this paper with the performance of an optimizer based on an exhaustive search strategy. The exhaustive optimizer is generated through Volcano Optimizer Generator (VOG). The results of the experiments indicate that the heuristic based optimizer has a superior performance with the increasing number of path expressions.