一种适应GPU的混合OLAP查询处理模型

发布时间：2018-11-12 11:14

【摘要】：通用GPU因其强大的并行计算能力成为新兴的高性能计算平台,并逐渐成为近年来学术界在高性能数据库实现技术领域的研究热点.但当前GPU数据库领域的研究沿袭的是ROLAP(relational OLAP)多维分析模型,研究主要集中在关系操作符在GPU平台上的算法实现和性能优化技术,以哈希连接的GPU并行算法研究为中心.GPU拥有数千个并行计算单元,但其逻辑控制单元较少,相对于CPU具有更强的并行计算能力,但逻辑控制和复杂内存管理能力较弱,因此并不适合需要复杂数据结构和复杂内存管理机制的内存数据库查询处理算法直接移植到GPU平台.提出了面向GPU向量计算特性的混合OLAP多维分析模型semi-MOLAP,将MOLAP(multidimensional OLAP)模型的直接数组访问和计算特性与ROLAP模型的存储效率结合在一起,实现了一个基于完全数组结构的GPU semi-MOLAP多维分析模型,简化了GPU数据管理,降低了GPU semi-MOLAP算法复杂度,提高了GPU semiMOLAP算法的代码执行率.同时,基于GPU和CPU计算的特点,将semi-MOLAP操作符拆分为CPU和GPU平台的协同计算,提高了CPU和GPU的利用率以及OLAP的查询整体性能.
[Abstract]:General GPU has become a new high performance computing platform because of its powerful parallel computing ability, and has gradually become a hot research topic in the field of high performance database implementation technology in academic circles in recent years. However, the current research in the field of GPU database is based on the ROLAP (relational OLAP) multidimensional analysis model. The research mainly focuses on the algorithm implementation and performance optimization technology of relational operators on GPU platform. GPU has thousands of parallel computing units, but its logical control units are few, and it has stronger parallel computing ability than CPU, but the ability of logic control and complex memory management is weak. Therefore, it is not suitable for the query processing algorithm of memory database which needs complex data structure and complex memory management mechanism to transplant directly to GPU platform. A hybrid OLAP multidimensional analysis model (semi-MOLAP,) for GPU vector computing is proposed, which combines the direct array access and computing characteristics of MOLAP (multidimensional OLAP) model with the storage efficiency of ROLAP model. A GPU semi-MOLAP multidimensional analysis model based on complete array structure is implemented, which simplifies GPU data management, reduces the complexity of GPU semi-MOLAP algorithm, and improves the code execution rate of GPU semiMOLAP algorithm. At the same time, based on the characteristics of GPU and CPU computing, the semi-MOLAP operator is divided into CPU and GPU platform, which improves the utilization of CPU and GPU and the overall performance of OLAP query.
【作者单位】：数据工程与知识工程教育部重点实验室(中国人民大学);中国人民大学信息学院;中国人民大学中国调查与数据中心;
【基金】：中央高校基本科研业务费专项资金(16XNLQ0,13XNLF01) 华为创新研究计划(HIRP 20140507,HIRP 20140510)~~
【分类号】：TP338.6

【参考文献】