【摘要】 索引作为加速数据库查询的一种成熟技术,始终受限于CPU的内存带宽与架构的发展,因此无法在性能上实现质的飞跃。所以使用GPU赋能索引技术来辅助数据库执行查询任务是势在必行的。因此,针对异构环境下索引结构的适应性以及现有GPU索引受限于显存容量导致扩展性不够等问题,提出了一种CPU与GPU协同处理的HPGB+-Tree索引算法。该算法以混合架构的方式重新构建索引结构,使其完全适应GPU的硬件特性,突破CPU内存带宽受限和GPU内存容量受限的双重难关。HPGB+-Tree索引不仅解决了索引异构问题,还充分利用两大硬件平台各自的优势加速基于索引的相关操作。在不同数据量与不同任务规模下对算法的性能进行了评估,实验结果表明,该算法在内核占用率与程序执行速度两个方面都极具优势,在性能上处于领先地位。更多还原

【Abstract】 As a mature technology to speed up database query,index is always limited by the development of CPU memory bandwidth and architecture,so it can not achieve a qualitative leap in performance. Therefore,it is imperative to use GPU enabled index technology to assist database query. Aiming at the adaptability of index structure in heterogeneous environment and the insufficient scalability of existing GPU index due to the limitation of video memory capacity,HPGB+-Tree index construction algorithm is proposed. The algorithm reconstructs the index structure in the way of hybrid architecture,makes it fully adapt to the hardware characteristics of GPU,and breaks through the dual difficulties of CPU memory bandwidth limitation and GPU memory capacity limitation. HPGB+-Tree not only solves the problem of index heterogeneity,but also makes full use of the advantages of the two hardware platforms to speed up the operation based on index. The performance of the algorithm is evaluated under different data volumes and different task sizes. The experimental results show that the algorithm has great advantages in both kernel occupancy and program execution speed,and the method is superior to other index algorithms,and HPGB+-Tree index is still in the leading position in performance.更多还原