电信科学 ›› 2023, Vol. 39 ›› Issue (9): 12-20.doi: 10.11959/j.issn.1000-0801.2023174

• 专题:网络智能化与生成式人工智能 • 上一篇    

算网资源智能适配与融合调度方法

张维庭, 孙呈蕙, 王洪超, 代嘉宁   

  1. 北京交通大学移动专用网络国家工程研究中心,北京 100044
  • 修回日期:2023-09-04 出版日期:2023-08-01 发布日期:2023-08-01
  • 作者简介:张维庭(1992- ),男,博士,北京交通大学移动专用网络国家工程研究中心副教授、硕士生导师,主要研究方向为工业互联网、算力网络和网络智能
    孙呈蕙(1998- ),女,北京交通大学移动专用网络国家工程研究中心硕士生,主要研究方向为算力网络和边缘计算
    王洪超(1982- ),男,博士,北京交通大学移动专用网络国家工程研究中心副教授、硕士生导师,主要研究方向为信息网络
    代嘉宁(2000- ),男,北京交通大学移动专用网络国家工程研究中心硕士生,主要研究方向为算力网络和深度强化学习
  • 基金资助:
    国家自然科学基金资助项目(62201029);中国博士后科学基金资助项目(2022M710007);中国博士后科学基金资助项目(BX20220029)

Intelligent adaptation and integrated scheduling method for computing and networking resources

Weiting ZHANG, Chenghui SUN, Hongchao WANG, Jianing DAI   

  1. National Engineering Research Center of Advanced Network Technologies, Beijing Jiaotong University, Beijing 100044, China
  • Revised:2023-09-04 Online:2023-08-01 Published:2023-08-01
  • Supported by:
    The National Natural Science Foundation of China(62201029);China Postdoctoral Science Foundation(2022M710007);China Postdoctoral Science Foundation(BX20220029)

摘要:

为有效支撑网络和计算深度融合的发展需求,新型的算力网络架构应运而生。在此背景下,如何实现算网资源的智能感知以及计算任务的高效调度,是当前网络需要解决的关键问题。为此,分析了面向算网融合的新型网络场景,设计了计算任务与算力节点的调度模型,提出了一种基于深度强化学习的资源调度算法。所提算法通过感知用户设备、算网资源可用容量和链路状态等关键信息,能够智能地做出系统成本最小的调度决策。最后,通过仿真实验验证了所提算法在节约系统成本方面的有效性。

关键词: 算力网络, 资源调度, 深度强化学习

Abstract:

To effectively support the development needs of the deep integration of networking and computing, a novel computing and network convergence architecture has emerged.In this context, how to realize the intelligent perception of computing and networking resources and the efficient scheduling of computational tasks are key problems.To this end, the new network scenario for computing and networking convergence were analyzed, a scheduling model for computing tasks and nodes was designed, and a deep reinforcement learning-based resource scheduling algorithm was proposed.The proposed algorithm was able to intelligently make scheduling decisions that minimize the system cost by sensing key information such as user devices, available capacity of computing and networking resources, and link status.Finally, the effectiveness of the proposed algorithm in saving system cost was verified by simulation experiments.

Key words: computing and network convergence, resource scheduling, deep reinforcement learning, The National Natural Science Foundation of China, China Postdoctoral Science Foundation

中图分类号: 

No Suggested Reading articles found!