数据与计算发展前沿 ›› 2022, Vol. 4 ›› Issue (5): 33-41.

CSTR: 32002.14.jfdc.CN10-1649/TP.2022.05.004

doi: 10.11871/jfdc.issn.2096-742X.2022.05.004

• 专刊:东数西算:开启算力经济时代的世纪工程(上) • 上一篇    下一篇

基于高性能计算环境的HPC算力编程模式

王小宁1,*(),卢莎莎1,吴璨1,和荣1,闫晓婷1,2,肖海力1,迟学斌1,2   

  1. 1.中国科学院计算机网络信息中心,北京 100083
    2.中国科学院大学,北京 100190
  • 收稿日期:2022-07-29 出版日期:2022-10-20 发布日期:2022-10-27
  • 通讯作者: 王小宁
  • 作者简介:王小宁：中国科学院计算机网络信息中心,副研究员,主要研究方向为网格计算、高性能计算、分布式计算与云服务。
    本文中主要承担工作为模型提炼和论文撰写。
    WANG Xiaoning is an associate professor at the Computer Network Information Center, Chinese Academy of Sciences. Her research interests include grid computing, high-performance computing, and cloud service.
    In this paper, she is responsible for HPC programming mod-eling and paper writing.
    E-mail: wxn@sccas.cn
  • 基金资助:
    中国科学院战略性先导科技专项项目（A类）(XDA19020101)

HPC Computing Power Programming Paradigm Based on CNGrid

WANG Xiaoning1,*(),LU Shasha1,WU Can1,He Rong1,YAN Xiaoting1,2,XIAO Haili1,CHI Xuebin1,2   

  1. 1. Computer Network Information Center, Chinese Academy of Sciences, Beijing 100083, China
    2. University of Chinese Academy of Sciences, Beijing 100190, China
  • Received:2022-07-29 Online:2022-10-20 Published:2022-10-27
  • Contact: WANG Xiaoning

RichHTML

0

PDF

221

摘要：

【目的】随着云计算、大数据、人工智能等技术的兴起和广泛应用,促进了基于多元算力的融合计算发展。在国家“东数西算”战略的指引下,充分发挥HPC算力优势,提供新型HPC算力编程模式,是新一代计算基础设施可编程能力的重要变革。【方法】分析了高性能计算环境服务模式发展和现有计算环境下不同的编程模式,提出了基于高性能计算环境的HPC算力编程模式HPC as a function,定义了HPC算力和任务模式的基本抽象,以及HPC算力编程模式的参考体系结构。【结果】HPC算力编程模式可支持科研业务中融合计算对HPC算力的基本需求,可将适用于HPC的计算任务分发到合适的计算资源执行并有效管理,相比传统的工作流系统提供了更大的灵活性和可编程性。【结论】HPC算力编程模式可望有效提高“东数西算”计算基础设施的可编程性。

关键词: 国家高性能计算环境CNGrid, CNgrid系统软件, 融合计算, 算力网络, 编程模式

Abstract:

[Objective] With the rise of new computing technologies including cloud computing, big data, and artificial intelligence, hybrid computing based on multiple computing power has been promoted. Under the guidance of the national strategy of “East-West Computing Requirement Transfer”, it is important to develop the programming paradigm for the new generation of computing infrastructure to take full advantages of the HPC computing power and provide a new HPC computing power programming model. [Methods] Firstly, the current computing service mode of the high-performance computing environment and the existing HPC programming paradigm are introduced. Secondly, an HPC com-puting power programming paradigm (HPC as a function) based on the high-performance computing environment is proposed according to the HPC computing power and task abstraction. Then design of the reference architecture of the HPC computing power programming paradigm is given. [Results] The HPC computing power programming paradigm can support the basic needs of scientific research for HPC computing power, and distribute tasks that are suitable for HPC to appropriate computing resources. Compared to a traditional workflow system, the new paradigm provides greater flexibility and program-mability. [Conclusions] The HPC computing power programming paradigm is expected to effectively improve the programmability of the new computing infrastructure.

Key words: CNGrid, CNGrid system software, hybrid computing, computing power network, programming paradigm