Estimating recursive query costs for various parallel environments

We introduce a recursive query evaluation method whose main goal is to obtain recursive query cost estimate without performing the query. In particular, we show that the execution estimate of the parallel processing of recursive queries can be done by exploiting the characteristics of the database profile during the evaluation process for request optimization. As many recursive queries involve computation of a transitive closure, the method estimates both transitive closure sizes and transitive closure execution costs. 1 Introduction To support such applications as integrated office systems, design and engeneering, or expert systems, database systems of the new generation must support complex objects and /or deduction as well as offer extensibility. In addition , these systems must deal with a wide range of computer architectures, and provide for parallelism and adaptability. Coupling the support of recursive queries on complex objects, the use of parallelism, extensiblity, and adaptability requires a new approach to database system design. A possible way to achieve this goal is to embed a large part of the coupling in a query-cost evaluation tool. This paper describes such a tool and focuses on cost evaluation techniques for complex recursive queries. ACE, an adaptative query-cost evaluator tool designed at the Bull Corporation Research Center [AND90a], provides extensiblity and adaptability through its library-oriented design. Four librairies provide parameters and formulas describing architectures, relational operations tool, it is possible to study the effect of chosen environment parameters on the cost of given queries, while keeping the values of the other environment parameters b e d. Thus, questions such as: " what is the best algorithm for a given parallel architecture ?", or "what is the best architecture for a given application (database + queries) ?", can be answered. ACE provides a means of experimentally evaluating new designs without having to actually build a DBMS to do so. For ACE to accept query trees involving a transitive closure operation it was necessary to define a methodology for the estimation of re-cursive query costs in the context of various parallel environments. Several efficient parallel transitive closure algorithms have been proposed to take advantage of parallel architectures. They can be classified in two classes: those algorithms that assume a shared memory architecture and those that assume a distributed memory architecture. In the first class, we mainly find parallel transitive closure algorithms as Warren and adaptations of Warren algorithms (for example [LU87]). The two main parallel …