数据与计算发展前沿 ›› 2022, Vol. 4 ›› Issue (6): 20-28.

CSTR: 32002.14.jfdc.CN10-1649/TP.2022.06.003

doi: 10.11871/jfdc.issn.2096-742X.2022.06.003

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

算力网络中基于算力标识的算力服务需求匹配

周舸帆(),雷波*()   

  1. 中国电信股份有限公司研究院,北京 102209
  • 收稿日期:2022-08-30 出版日期:2022-12-20 发布日期:2022-12-20
  • 通讯作者: 雷波
  • 作者简介:周舸帆,中国电信股份有限公司研究院,工程师。主要研究方向为算力网络、算力标识与未来网络技术等。
    本文中负责算力服务需求匹配系统架构与整体方案设计。
    ZHOU Gefan is a researcher and eng-ineer at the Network Technology Rese-arch Department, China Telecom Co., Ltd. Research Institute. Her research interests include Computing Power Network, Computing power identification, future network technology, etc.
    In this paper, she is responsible for the overall architecture and scheme design of the computing power requirement matching system.
    E-mail: zhougf@chinatelecom.cn|雷波,中国电信股份有限公司研究院网络技术研究所,副所长,正高级工程师。中国科学院计算机网络信息中心客座研究员,北京邮电大学兼职教授,并兼任中国通信标准化协会CCSA TC617边缘计算产业发展及标准推进委员会副主席,CCSA TC614网络5.0算力网络特别工作组副组长。目前聚焦在云网融合、算力网络、未来网络技术等领域的研究。
    本文中负责指导系统研究与实现途径。
    LEI Bo, a professor-level senior engineer, is the deputy director of the Network Technology Research Department, China Telecom Co., Ltd. Research Institute. He currently works as visiting research fellow of Computer Network Information Center, Chinese Academy of Sciences, a part-time professor at the Beijing University of Posts and Telecommunications, vice-chairman of CCSA TC617 “Edge Computing Industry Development and Standards Promotion Committee”, the deputy leader of Computing power network special work group in CCSA TC614 “Network 5.0 Technical Standards Promotion Committee”. His research interests include cloud-network convergence, Computing Power Network, future network technology, etc.
    In this paper, he is responsible for guiding the research content and implementation methods of the system.
    E-mail: leibo@chinatelecom.cn
  • 基金资助:
    国家重点研发计划(2021YFB2900200)

Computing Service Demand Matching Based on Computing Power Identification in Computing Power Network

ZHOU Gefan(),LEI Bo*()   

  1. Research institute of China Telecom, Beijing 102209, China
  • Received:2022-08-30 Online:2022-12-20 Published:2022-12-20
  • Contact: LEI Bo

摘要:

【目的】随着数字经济发展不断深入与新型行业应用场景的不断涌现,全社会算力需求持续增长,“东数西算”等国家战略的逐步落地为以算力网络为代表的新型网络技术的发展带来强劲的驱动力。目前网络中的算力资源呈现多级泛在归属复杂的特点,因此针对多样化算力资源的算力需求匹配成为重要的研究方向。【方法】本文提出了一种基于算力标识的算力服务需求匹配方法,该方案引入了一种以算网融合为基础的算力标识方法,通过算力资源通信地址的分级映射、算力资源信息获取与算网信息融合实现算力服务需求的匹配。【结果】该方案通过结合算力资源通信属性和算力资源计算属性,可依据算力使用者需求匹配最佳算力,同时算力标识可根据实际情况进行弹性化的描述,具有良好的可扩展性与通用性。【局限】鉴于算力资源的多样性与复杂性,算力度量与算力标识的研究尚未成熟,未来随着学术与标准化工作的推进,将不断地发展与完善。【结论】算力标识的设计基于算力资源的特点,能较好地对算力资源属性进行描述,将用于对整网算力资源进行管理,同时实现算力服务需求的高效匹配,为算力资源调度与算网一体化管控提供基础。

关键词: 算力网络, 算网融合, 算力服务, 算力标识

Abstract:

[Objective] With the development of the digital economy and the emergence of new industry application scenarios, the demand for computing resources in the whole society grows continually. The gradual implementation of several national strategies such as the East-West Computing Requirement Transfer project has been promoting the development of advanced network technologies represented by the computing power network (CPN). At present, the computing resources in the network are ubiquitous, belonging to various owners. Therefore, the matching of computing resource demand for diverse computing resources has become an important research direction. [Methods] This paper presents a method for the matching of computing service demand based on computing power identification and a unique method of computing power identification based on computing-network convergence, where the matching of computing service demand is achieved by using the hierarchical mapping of the communication address of the computing resources, the acquisition of computing resource information and the combined information of computing and network. [Results] By combining communication properties and computing properties of the computing resources, the system matches the optimal computing resources for the computing resource consumers. The computing power identification is flexible and universal in describing the properties of the computing resources comprehensively. [Limitations] The research on computing power measurement and computing power identification is at the early stage and facing the complexity and diversity of computing resources. In the future, it can be further improved with the progress in research and standardization work. [Conclusions] Computing power identification proposed in this paper can completely describes the properties of the computing resources and achieve management of the computing resources in the network and the matching of computing service demand, which is the basis of computing power scheduling and joint management of computing and network.

Key words: computing power network, computing-network convergence, computing service, computing power identification