摘要
新兴的无服务边缘计算(Serverless Edge Computing,SEC)可在降低任务计算延迟的同时高效应对多变的服务模式,进而提升资源利用效率.然而,在资源受限的SEC中部署服务时常出现频繁的镜像置换,导致了过度的服务延迟与通信成本.现有的解决方案通常通过修改镜像结构或移动仓库位置来降低镜像请求时间,但其在一定程度上违背了容器隔离的设计初衷,并造成了额外的计算与存储开销.为了解决这些重要挑战,本文提出了一种新颖的面向SEC环境的镜像缓存与资源分配(Image Caching and Resource Allocation,ICRA)框架,并将原问题解耦为两个子问题分别进行求解.针对镜像缓存子问题,设计了一种基于改进深度强化学习(Deep Reinforcement Learning,DRL)的镜像缓存方法,通过引入双critic网络与延迟更新机制,以提升镜像缓存性能.针对资源分配子问题,根据任务属性与队列负载,引入凸优化理论进行容器资源分配以降低任务完成延迟.大量实验验证了所提ICRA框架的有效性.与基准方法相比,ICRA框架能够在保证服务质量的同时显著降低系统成本,并在不同场景下均表现出更加优越的性能.
The emerging Serverless Edge Computing(SEC)can efficiently handle variable service patterns while reducing task computation latency,thereby enhancing resource utilization efficiency.However,frequent image swapping often happens when deploying services in resource-constrained SEC environments,leading to excessive service delay and communication costs.Existing solutions typically mitigate image request time by modifying image structures or relocating repositories,which partially violate the original design principle of container isolation and incur extra computational and storage overheads.To address these important challenges,this paper proposes a novel Image Caching and Resource Allocation(ICRA)framework for SEC environments,decomposing the original problem into two subproblems to solve separately.For the subproblem of image caching,an improved Deep Reinforcement Learning(DRL)based image caching method is designed,incorporating twin critics′networks and delayed update mechanisms to enhance image caching performance.For the subproblem of resource allocation,the convex optimization theory is introduced for container resource allocation based on task attributes and queue loads to reduce task completion delay.Extensive experimental evaluations have confirmed the efficacy of the introduced ICRA framework.In comparison with benchmark methods,the ICRA framework significantly reduces system costs while ensuring service quality,which exhibits superior performance in various scenarios.
作者
曾旺
池梦莉
于正欣
苗旺
陈哲毅
ZENG Wang;CHI Mengli;YU Zhengxin;MIAO Wang;CHEN Zheyi(College of Computer and Data Science,Fuzhou University,Fuzhou 350116,China;Key Laboratory of Spatial Data Mining&Information Sharing,Ministry of Education,Fuzhou 350002,China;Fujian Provincial Key Laboratory of Networking Computing and Intelligent Information Processing,Fuzhou 350116,China;School of Computing and Communications,Lancaster University,Lancaster LA14YW,UK;Department of Computer Science,University of Exeter,Exeter EX44QF,UK)
出处
《小型微型计算机系统》
北大核心
2025年第10期2515-2522,共8页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(62202103)资助
中央引导地方科技发展资金项目(2022L3004)资助
福建省科技经济融合服务平台项目(2023XRH001)资助
福厦泉国家自主创新示范区协同创新平台项目(2022FX5)资助.
关键词
无服务边缘计算
镜像缓存
资源分配
深度强化学习
凸优化
serverless edge computing
image caching
resource allocation
deep reinforcement learning
convex optimization
