Spatial query processing engine in spatially enabled database

In spatially enabled database, spatial data and traditional structured data share same query processing engine (QPE) through the standard SQL interface. It\´s essential to design a robust QPE to efficiently support and process query workflow which is mixed with spatial and non-spatial predictions. Currently most of spatially enabled databases adopt the "blackbox model" to support the spatial query processing. However, due to lack of the mechanism in spatial data type and prediction, current approaches are still difficult to generate efficient spatial query execution plans. Our paper firstly proposes an multi-layer mechanism to extend spatial data type and function for the QPE. QPE can add or remove spatial modules without affecting the rest of the system. Secondly a cost evaluation framework to process different query flow is introduced. The framework calculates each query plan cost and chooses the optimal query plan to execute in bottom-up fashion. The cost evaluation in QPE is aware of spatial operations. The interfaces of optimizer in QPE are designed to improve the accuracy of cost parameters. Considering the importance of cardinality estimation in cost calculation, an revised probability cumulative density histogram are introduced to generalize spatial data distribution and improve the accurate selectivity and cardinality estimation of spatial selection in cost calculation phase. This histogram uses probability density to interpolate cells which intersect with regular cells in any corner points or edges. Meanwhile, statistics are collected from the registered tables which monitor the in-memory statement of different components and then feed back to optimizer in QPE to activate inter-query optimization. Thirdly, we also introduce spatial join physical operators into query execution component to enhance spatial join execution efficiency. Our spatially enabled database and QPE prototype are implemented on open source Ingres database and compatible with SQL/MM spec- - ification. Real datasets are used to verify QPE correctness and efficiency. Our experiments show that the QPE improves the accuracy of spatial predictions in cost estimation. The average error of spatial selection is less than 20%. Estimated cost for each query execution plan is positively correlative with actual execution time. And the introduced spatial physical operators in execution component improve more than 10% execution efficiency in the optimal query plan.