华东师范大学学报(自然科学版) ›› 2014, Vol. 2014 ›› Issue (5): 240-251.doi: 10.3969/j.issn.10005641.2014.05.021

• 计算机科学与技术 • 上一篇    下一篇

Co-OLAP: CPU&GPU混合平台上面向星形模型基准的协同OLAP

张宇1,2，张延松1,2,3，张兵1,2，陈红1,2，王珊1,2   

  1. 1. 中国人民大学 DEKE实验室， 北京 100872； 2. 中国人民大学 信息学院，北京 100872； 3. 中国人民大学 中国调查与数据中心，北京 100872
  • 出版日期:2014-09-25 发布日期:2014-11-27
  • 通讯作者: 张延松，男，博士后，讲师，研究方向为内存数据库、OLAP E-mail:zhangys_ruc@hotmail.com
  • 作者简介:张宇，女，博士研究生，研究方向为GPU、OLAP. Email: zyszy511@hotmail.com.
  • 基金资助:

    中央高校基本科研业务费专项资金（12XNQ072, 13XNLF01）

Co-OLAP: Research on cooperated OLAP with star schema benchmark on hybrid CPU&GPU platform

 ZHANG  Yu1,2, ZHANG  Yan-Song1,2,3, ZHANG  Bing1,2, CHEN  Hong1,2, WANG  Shan1,2   

  1. 1. DEKE Lab, Renmin University of China, Beijing 100872, China;
    2. School of Information, Renmin University of China, Beijing 100872, China;
    3. National Survey Research Center at Renmin University of China, Beijing 100872, China
  • Online:2014-09-25 Published:2014-11-27

PDF

1383

摘要： 当前GPU(图形处理器),即使是中端服务器配置的中端GPU也拥有强大的并行计算能力.不同于近期的研究成果,中端服务器可能配置有几块高端CPU和一块中端GPU,GPU能够提供额外的计算能力而不是提供比CPU更加强大的计算能力.本文以中端工作站上的Co-OLAP(协同OLAP)为中心,描述如何使中端GPU与强大的CPU协同以及如何在计算均衡的异构平台上分布数据和计算以使Co-OLAP模型简单而高效.根据实际的配置,基于内存容量,GPU显存容量,数据集模式和订制的AIR(数组地址引用)算法提出了最大高性能数据分布模型. Co-OLAP模型将数据划分为驻留于内存和GPU显存的数据集,OLAP计算也划分为CPU和GPU端的自适应计算负载来最小化CPU和GPU内存之间的数据传输代价.实验结果显示,在SF=20的SSB(星形模型基准)测试中,两块至强六核处理器的性能略优于一块NVIDA Quadra 5 000 GPU(352个cuda核心)的处理性能, Co-OLAP模型可以将负载均衡分布在异构计算平台并使每个平台简单而高效.

关键词: GPU(图形处理器), OLAP(联机分析处理), Co-OLAP(协同OLAP), AIR(数组地址引用)

Abstract: Nowadays GPUs have powerful parallel computing capability even for moderate GPUs on moderate servers. Opposite to the recent research efforts, a moderate server may be equipped with several high level CPUs and a moderate GPU, which can provide additional computing power instead of more powerful CPU computing. In this paper, we focus on Co-OLAP(Cooperated OLAP) processing on a moderate workstation to illustrate how to make a moderate GPU cooperate with powerful CPUs and how to distribute data and computation between the balanced computing platforms to create a simple and efficient Co-OLAP model. According to real world configuration, we propose a maximal high performance data distribution model based on RAM size, GPU device memory size, dataset schema and special designed AIR(array index referencing) algorithm. The Co-OLAP model distributes dataset into host and device memory resident datasets, the OLAP is also divided into CPU and GPU adaptive computing to minimize data movement between CPU and GPU memories. The experimental results show that two Xeon six-core CPUs slightly outperform one NVIDA Quadra 5 000 GPU with 352 cuda cores with SF=20 SSB dataset, the Co-OLAP model can assign balanced workload and make each platform simple and efficient.

Key words: GPU, OLAP, Co-OLAP, AIR

中图分类号: