作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (6): 52-59. doi: 10.19678/j.issn.1000-3428.0057965

• 人工智能与模式识别 • 上一篇    下一篇

图数据库中基于GPU的图分析计算方法

钱裳云1,2,3, 邵志远1,2,3, 郑然1,2,3, 陈继林4   

  1. 1. 华中科技大学 计算机科学与技术学院, 武汉 430074;
    2. 华中科技大学 服务计算技术与系统教育部重点实验室, 武汉 430074;
    3. 华中科技大学 集群与网格计算湖北省重点实验室, 武汉 430074;
    4. 中国电力科学研究院有限公司, 北京 100192
  • 收稿日期:2020-04-03 修回日期:2020-05-20 发布日期:2020-06-15
  • 作者简介:钱裳云(1996-),女,硕士研究生,主研方向为图计算;邵志远、郑然,副教授;陈继林,高级工程师。

GPU-based Graph Analysis and Computation Method for Graph Database

QIAN Shangyun1,2,3, SHAO Zhiyuan1,2,3, ZHENG Ran1,2,3, CHEN Jilin4   

  1. 1. College of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan 430074, China;
    2. Key Laboratory of Services Computing Technology and System of Minstry of Education, Huazhong University of Science and Technology, Wuhan 430074, China;
    3. Key Laboratory of Cluster and Grid Computing in Hubei province, Huazhong University of Science and Technology, Wuhan 430074, China;
    4. China Electric Power Research Institute, Beijing 100192, China
  • Received:2020-04-03 Revised:2020-05-20 Published:2020-06-15
  • Contact: 国家电网有限公司总部科技项目“适用于电力系统应用的高性能计算技术研究与开发”(XTB17201900305)。 E-mail:zyshao@hust.edu.cn

摘要: 现有的图数据库对于在线分析操作大多采用基于CPU的分布式图计算引擎(如GraphX),但CPU核心数量有限的不足会导致计算效率低下,同时集群间的同步也会产生额外的通信开销。通过使用图形处理单元(GPU)对图计算进行加速,设计并实现图处理系统RockGraph。该系统能够根据用户需求从图数据库中提取出包含核心信息的子图,经过数据格式转换后,利用JNI工具调用动态链接库,采用超显存GPU图计算框架进行在线分析,并将计算结果写回图数据库。实验结果表明,与基于CPU的分布式图计算系统相比,RockGraph的图分析效率可提高3倍~5倍。

关键词: 图数据库, 图分析计算, 图形处理单元, 子图提取, 超显存计算

Abstract: Most of the existing graph databases use CPU-based distributed graph computing engines(such as GraphX) for online analysis operations, but the limited number of CPU cores will lead to low computing efficiency, and the synchronization between clusters will also generate additional communication overhead.This paper presents the design and implementation of a graph processing system, RockGraph, which uses Graphics Processing Units(GPU) to accelerate the graph computation.The system extracts the subgraph containing core information from the graph database according to the user needs.After the data format conversion, the dynamic link library is called by using the JNI tool, and the GPU graph computing framework is used for online analysis.Then the calculation results are written back to the graph database.Experimental results show that, compared with the CPU-based distributed graph computing system, the proposed RockGraph system increases the graph analysis efficiency by 3 times to 5 times.

Key words: graph database, graph analysis and computation, Graphics Processing Unit(GPU), subgraph extraction, out-of-memory computation

中图分类号: