Performance evaluation of catalog architectures in distributed database systems

Abstract

Catalog architectures affect many aspects of database systems such as site autonomy, query optimization, view management, authorization mechanism, and data distribution transparency. However, the performance of various catalog architectures has received relatively little attention. Embedded read queries to the catalogs(data manipulation statements) assumed to be compiled into an access module, and the module executed repeatedly later or immediately with different parameters. Update queries to the catalogs(data definition and authorization statements) are assumed to be interpreted due to their interactive nature, and actual updates to the catalogs are performed. The performance of six catalog architectures from the viewpoint of query compilation (interpretation) is investigated first using simulation approach both in a locally distributed database system and a geographically distributed database system. The six schemes studied include a centralized catalogs, two variations of fully replicated catalogs, and three variations of partitioned catalogs. We show that the three variations of partitioned catalogs perform better than the centralized catalogs and two variations of fully replicated catalogs under almost simulation conditions. The centralized catalogs and the fully replicated catalogs with quorum consensus give the worst performance due to the queuing delay in several queues represented in the simulation model. The results of this study also indicate that the performance difference among the variations of the partitioned catalogs is mainly due to the different recompilation rates and disk access requirements. In order to obtain the performance of the catalog architectures from yet another viewpoint of running access module, the three catalog architecture alternatives-the centralized catalogs, the fully replicated replicas with unanimous agreement, and the partitioned catalogs-are examined and analyzed using an extended simulation model. Among the three catal...