TOPAZ: a Cost-Based, Rule-Driven, Multi-Phase Parallelizer

Currently the key problems of query optimization are extensibility imposed by object-relational technology, as well as query complexity caused by forthcoming applications, such as OLAP. We propose a generic approach to parallelization, called TOPAZ. Different forms of parallelism are exploited to obtain maximum speedup combined with lowest resource consumption. The necessary abstractions w.r.t. operator characteristics and system architecture are provided by rules that are used by a cost-based, top-down search engine. A multi-phase pruning based on a global analysis of the plan efficiently guides the search process, thus considerably reducing complexity and achieving optimization performance. Since TOPAZ solely relies on the widespread concepts of iterators and data rivers common to (parallel) execution models, it fits as an enabling technology into most state-of-the-art (object-) relational systems.