关键词: 列数据库，调度方法，异构计算

Abstract: The heterogeneous computing platforms composed of mufti-core CPU and GPU have been a important devc lopment direction of current high performance computing. In order to improve the performance of column-base database query effectively and make full use of the computing resources of heterogeneous computing platforms,based on a set of defined columrrbase database primi-lives, this paper presented a set of primitive schedule methods, which include a primi- live execution mechanism, a CPU primitive schedule method based on dynamic programming and a GPU primitive schedule method based on GPU memory management mechanism. It maks that system can reasonably make use of mufti core CPU computing resources, effectively make use of locality of data in GPU memory and improve the overall performance. This paper tested several typical queries in TPGH benchmark programs, and the result represents that the CPU primitive schedule method lets the performance of queries more stable and GPU schedule method makes it faster. Meanwhile, through the experiments, we found that these column-base database schedule methods on heterogeneous computing platforms have some defects, which indicate the improvement direction for follow-up work.

Key words: Column-based database, Schedule method, Heterogeneous computing