Query optimizers for persistent object systems should be extensible to react to user-supplied abstract types. Current architectures support only a single, non-extensible technique for controlling the optimization process. The authors propose an alternative to the current extensible architectures that will support multiple optimizer control strategies and the addition of new control strategies. The optimizer consists of a collection of optimization regions, each of which can transform queries according to a particular control strategy, set of transformations and cost model. A global optimizer control coordinates the movement of a query between these regions. This architecture provides extensibility in the optimizer's repertoire of control strategies through the addition of new regions. The authors describe their approach and demonstrate its utility by following the optimizer as it works on an example query. The optimizer will move the query between three distinct regions. The different regions illustrate different kinds of transformations and different strategies for application of those transformations.
展开▼
