How to improve performance of complex database query is an important topic of database research. We introduce a new column database query engine, which is based on the vector-primitive operation set, to improve the database query performance by exploiting primitive-level parallelism on the multi-core processors. We present the engine's core components, including the vector primitives, the storage layout of the column-stored data file, vector primitive dependence graph for complex query, the dynamic scheduler and memory management. According to the experiments on four typical queries in TPC-H, it is found that complex database query is computing-intensive application and vector primitives can be executed in parallel. The result shows that, on multi-core processor system, our engine's query performance is 1. 2 times of which on single work-thread and 20 times of MySQL.%提升复杂数据库查询性能是数据库研究的重要方向.提出了一个基于向量原语操作的列数据库查询引擎,通过在多核处理器上开发向量原语之间的并行性来提高数据库查询的性能.介绍了该查询引擎中的向量原语设计、列数据文件存储布局、查询的向量操作依赖图、向量操作动态调度器和内存管理器等关键部件.对于TPC-H 4个典型复杂查询的测试结果表明:1)数据库复杂查询是典型的计算密集型操作,而且原语之间存在着一定的可并行成分；2)在多核处理器上,该查询引擎可以有效开发向量原语间操作的并行性,性能较单线程执行可以提升1.2倍左右,较MySQL则可以最高提升20倍.
展开▼
