pdf (870 Kb) - Fachgebiet Datenbanken und Informationssysteme

Abstract
The aim of the query evaluation in relational data base systems consists of generating
the optimal execution plan for a given query. Since this is generally a
very complex matter, one reduces the requirements in practice to avoid bad query
trees.
This diploma presents an introduction to this complex topic, in order to examine
afterwards the heuristic query optimization closer. By giving a basic cost model,
it gives one the possibility, to make relative statements about the quality of the
optimized query plan. Not the exactness of the costs is important, but the relative
cost differences of the respective query plans.
These alternatives are developed in two following phases of the optimization.
First an algebraic optimization is made, with which equivalence rewritings are
accomplished by transformation rules. Since for a logical algebra operator often
different implementations exist, the most economical plan is selected in the following
physical optimization. In both cases heuristics are used for the controlling
of generating alternatives. They represent empirical values for the meaningful
application of certain rewriting rules.
The search methods needed for the query optimization offer afterwards the possibility
of improving the developed query tree. The additional given metadatas
permit to use a selection of search methods, which can find a more economical -
again accelerated by heuristics - query.
For the conclusion of this work the presented query optimizer is simulated, in
order to make closer investigations possible concerning the fine tunings. Special
attention is directed on the necessary flexibility and the expandability of the
programmed optimization modules.