多层次算力网络下的资源管控优化技术  

Resource Management and Optimization Technology Under Multi-level Computing Networks

作　　者：周宗杰 赵哲[1] 宋西斐 黄航 冯杰 ZHOU Zongjie;ZHAO Zhe;SONG Xifei;HUANG Hang;FENG jie(ShanXi Key Laboratory of Blockchain and Secure Computing,Xidian University,Xi’an 710071,China)

机构地区：[1]西安电子科技大学陕西省区块链与安全计算重点实验室,陕西西安710071

出　　处：《移动通信》2025年第3期86-91,138,共7页Mobile Communications

基　　金：中国博士后科学基金面上项目“边缘智能驱动的多层次网络多维资源管控技术”(2024M752525)。

摘　　要：算力网络通过整合分布式计算资源,提升计算效率、降低能耗并支持多样化需求,受到了广泛关注。阐述了算力网络的发展现状及其面临的技术挑战,提出了一种端-边-云协同算力网络架构,实现了分布式算力资源的高效管理。为优化多层次资源分配,提出基于IMPALA的资源调度算法。实验结果表明提出的算法和基线算法相比能够更有效地优化网络资源,降低网络延迟。上述研究对未来算力网络资源的高效管理具有一定的指导意义。The computing power network,by integrating distributed computing resources,has garnered widespread attention for improving computational efficiency,reducing energy consumption,and supporting diverse needs.This paper discusses the current development status of computing power networks and the technical challenges they face,proposing an end-edge-cloud collaborative computing power network architecture that achieves efficient management of distributed computing resources.To optimize multi-level resource allocation,a resource scheduling algorithm based on IMPALA(Importance Weighted Actor-Learner Architecture)is proposed.Experimental results show that the proposed algorithm can more effectively optimize network resources and reduce network latency compared to baseline algorithms.The above research provides significant guidance for the efficient management of future computing power network resources.

关 键 词：算力网络 IMPALA算法 资源调度 

分 类 号：TN92[电子电信—通信与信息系统]