面向CUDA的GPU虚拟化的研究及应用.pdfVIP

 面向CUDA 的GPU 虚拟化研究及其应用 + 马业 , 袁家斌, 吕相文 (南京航空航天大学 计算机科学与技术学院,江苏 南京 210016) * Research and Application of CUDA-based GPU Virtualization MA Ye, YUAN Jia-Bin, LV Xiang-Wen (College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China) + Corresponding author: E-mail: maye1766@126.com, http://www.nuaa ./nuaanew Abstract : Considering the powerful parallel computing ability of GPU and the isolation characteristic of virtualization,the GPU virtualization technology that support dynamic scheduling and multi -user concurrent is presented. CUDA manage end utilize centralized, flexible mechanism to manage the GPU resources. Simultaneously,CUDA manage end manage GPU internal tasks. We set a value of integrated load evaluation to achieve load balance.We design a distributed heterogeneous system to simulate the application scenarios.At the same time,We verify the feasibility and efficiency of GPU virtualization that applied in the designed system. Key words: GPU virtualization; CUDA; distributed system 摘 要: 为高效、充分的利用GPU，针对GPU 的特点，结合虚拟化环境的安全、隔离等特性，借鉴已有的虚拟 化技术，提出了一种可动态调度、支持多用户并发的 GPU 虚拟化技术。CUDA 管理端采用集中、灵活的机制对 GPU 资源进行统一管理，对GPU 内部任务统一调度。GPU 内部任务调度通过设置综合负载评价值实现负载平衡， 避免了资源的利用不充分。设计了一个面向科学计算的分布式异构系统来模拟 GPU 虚拟化的应用场景，并通过 在设计的系统上进行大规模矩阵运算实验，验证了GPU 虚拟化技术应用在计算系统中的可行性和高效性。 关键词: GPU 虚拟化；CUDA ；分布式系统 中图法分类号: TP393 文献标识码: A  Supported by the National High-Tech Research and Development Plan of China under Grant No.2009AA044601( 国家高技术研究 发展计划(863)); the National Natural Science Foundation of China under Grant No 国家自然科学基金重点项目); A Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutio