内嵌式空间数据库优化器代价评估框架及实现

Optimizer Cost Estimation Framework and Implementation for Spatially-Enabled Database

传统的优化器缺乏空间数据分布统计信息和空间算子的支持,不能评估和生成有效的查询执行计划。对此,提出了包含空间执行算子的一体化评估框架,根据重要属性及元组流属性实现了执行计划自底向上的代价推演计算。通过服务器端编程实现了利用概率累计扩展空间直方图模型描述空间数据的统计信息,并通过优化器接口计算空间谓词选择率,为优化器提供更为准确的代价参数,改进了评估的准确度。实验结果表明,采用此方法进行涉及空间查询的计算,理论代价估计与实际执行代价具有相对一致性,在统计值完整的情况下能够准确地估算实际较优的执行计划。

Query optimizer is critical component in spatially enabled databases.Processing complex spatial and non-spatial queries in seamless fashion is responsiple.However,the lack of spatial statistics and spatial operators limits the accuracy of cost derivation and the generation of optimal query execution plan.We propose an integrated cost evaluation framework in optimizer.Based on the important keys of each node and the characteristic of pipeline,the cost of query execution plan is derived bottom-up.An improved cumulative density histogram is introduced to describe the distribution of spatial data,and is implemented as a module in database server through extensible server-side programming interface.Optimizer uses these statistics to improve the accuracy of cardinality estimation.The experiments illustrate that the cost evaluation framework can correctly and efficiently evaluate different execution plans.The results of different hybrid query statements show the positive correlation between estimated cost and actual cost.