算网自智:研究进展与展望  

Computing-network intelligence:research progress and prospects

作　　者：朱海龙 杨帆[1] 蒋如一 贾庆民 周晓茂 谢人超[1,2] 黄韬 ZHU Hailong;YANG Fan;JIANG Ruyi;JIA Qingmin;ZHOU Xiaomao;XIE Renchao;HUANG Tao(State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China;Purple Mountain Laboratories,Nanjing 211111,China)

机构地区：[1]北京邮电大学网络与交换技术国家重点实验室,北京100876 [2]紫金山实验室,江苏南京211111

出　　处：《通信学报》2024年第10期191-206,共16页Journal on Communications

基　　金：国家自然科学基金资助项目(No.92267301,No.92367104)。

摘　　要：尽管算力网络可以通过网络连接并灵活地调度计算资源,但在实际落地应用中仍然缺乏智能性和灵活性。针对算力网络亟须提高智能化水平的问题,首先简要引出了算网自智的背景和概念;接着提出了一种新型的算网自智基本架构和工作机制,通过算网感知、智能决策、算力路由、自动化和自优化等关键技术实现算力网络多维资源的智能资源分配和服务调度;最后,对算网自智未来的主要研究方向和技术挑战作了展望。算网自智作为未来的必然发展趋势,可以很好地解决运维流程复杂、过度依赖人工经验、难以自适应动态变化的环境等一系列问题。Although the computing power network(CPN)can connect via the Internet and flexibly schedule computing resources,it still lacks intelligence and flexibility in practical applications.To address the need for enhanced intelligence in computing power networks,the background and concept of the computing-network intelligence(CNI)were first introduced briefly.Then,followed by the proposal of a novel basic architecture and operational mechanism for CNI,intelligent resource allocation and service scheduling of multi-dimensional resources were achieved through key technologies such as computing-network perception,intelligent decision-making,computing power routing,automation,and selfoptimization.Finally,future research directions and technical challenges were discussed.CNI is recognized as an inevitable trend for the future,effectively addressing issues such as complex operational processes,excessive reliance on human experience,and difficulties in adapting to dynamically changing environments.

关 键 词：算网融合 算网自智 算力网络 自智网络 

分 类 号：TN91[电子电信—通信与信息系统]