摘要： 叙述了在扩展BSP计算机上实现的shared_nothing 并行对象关系数据库（PORDB）的两阶段查询优化。它分为顺序优化阶段和并行化两个阶段。顺序优化阶段在预测并行通信代价的基础上，采用基于代价的顺序查询优化。并行化阶段为缩减搜索空间，先按照并行模式划分并行单元，再以并行单元为单位分配处理机。BSP模型可简化并行模式的实现和并行单元的划分。

关键词: 并行对象关系数据库, BSP计算机, 两阶段查询优化, 并行单元, 并行模式

Abstract: Based on extended BSP computer, we implemented two-phrase query optimization of shared_nothing parallel object-relational database(PORDB). Two-phrase optimization includes sequential optimization and parallelization. Sequential phrase uses sequential optimization based on cost . In order to cut down the enormous search space, parallelization first divides query tree into parallel units according to parallel forms, then allocates processes to parallel units. The BSP model simplized the implemention of parallel forms and the division of parallel units.

中图分类号: 

• TP311