Parallel query evaluation in deductive databases

Won Suk Lee, Purdue University

Abstract

This research investigates a set of query evaluation schemes in deductive databases on parallel computing environments. Based on two different parallel environments: massively parallel environment and conventional multiprocessor environment, query evaluation schemes are studied. A deductive database can be implemented as a collection of objects which forms a connected network of a large number of simple processing elements. Data objects can be actively involved in the evaluation process of a query. With this approach, the tuple-by-tuple, operation-by-operation evaluation strategy employed by most database systems can be avoided. In addition, the need to maintain temporary relations can be avoided. For a conventional multiprocessor environment, a multiple query evaluation scheme is proposed. When multiple queries can be evaluated together, some queries may share certain database operations. In order to identify the common database operations in a set of queries, they are compiled into a network called a Relational Network which eliminates redundant database operations and allows multiple operations to be performed simultaneously. To extract more parallelism, relations in a database are partitioned horizontally and vertically. In addition, vector notations are introduced to eliminate the need of constructing temporary relations, to reduce the amount of data transfer among a set of processors, and to allow set operations to be performed without actually accessing the data. Finally, one of the most important problems in implementing a deductive database is that of processing logical transactions efficiently. Considering only the sharing of common database operations among the queries, the number of database operations can be decreased. However, fewer number of database operations cannot guarantee lower evaluation cost. Consequently, enhancing a set of conjunctive queries can be achieved by considering the sharing of common database operations and the ordering of conjuncts in each query, so that the total evaluation cost can be decreased. Based on the sharing and ordering of conjunctive queries, a set of query optimization strategies are discussed in this research. (Abstract shortened with permission of author.)

Degree

Ph.D.

Advisors

Sheu, Purdue University.

Subject Area

Computer science

COinS