A query processing system for very large spatial databases using a new map algebra.

摘要

In this thesis we introduce a query processing approach for spatial databases and explain the main concepts we defined and developed: a spatial algebra and a graph based approach used in the optimizer. The spatial algebra was defined to express queries and transformation rules during different steps of the query optimization. To cover a vast variety of potential applications, we tried to define the algebra as complete as possible. The algebra looks at the spatial data as maps of spatial objects. The algebraic operators act on the maps and result in new maps. Aggregate functions can act on maps and objects and produce objects or basic values (characters, numbers, etc.). The optimizer receives the query in algebraic expression and produces one efficient QEP (Query Evaluation Plan) through two main consecutive blocks: QEG (Query Evaluation Graph) generation and QEP generation. In QEG generation we construct a graph equivalent of the algebraic expression and then apply graph transformation rules to produce one efficient QEG. In QEP generation we receive the efficient QEG and do predicate ordering and approximation and then generate the efficient QEP. The QEP is a set of consecutive phases that must be executed in the specified order. Each phase consist of one or more primitive operations. All primitive operations that are in the same phase can be executed in parallel.; We implemented the optimizer, a randomly spatial query generator and a simulated spatial database. The query generator produces random queries for the purpose of testing the optimizer. The simulated spatial database is a set of functions to simulate primitive spatial operations. They return the cost of the corresponding primitive operation according to input parameters. We put randomly generated queries to the optimizer, got the generated QEPs and put them to the spatial database simulator. We used the experimental results to discuss on the optimizer characteristics and performance.; The optimizer was designed for databases with a very large number of spatial objects nevertheless most of the concepts we used can be applied to all spatial information systems. (Abstract shortened by UMI.)