摘要
为充分利用边缘服务器的有限资源,提高应用服务的缓存效益,本文提出了以应用服务缓存为基础的协同卸载的车载边缘计算模型。在此基础上,以卸载任务的时延和能耗最小化为优化目标,展开对应用服务缓存和计算卸载问题的研究。将服务缓存、任务卸载以及计算资源分配的联合优化建模为非线性整数规划问题。为降低求解难度,将原问题分解为服务缓存和计算卸载联合决策子问题以及边缘服务器计算资源分配子问题。其中,将服务缓存和计算卸载子问题建模为马尔科夫决策过程,并提出了一种基于深度强化学习的缓存卸载方案。仿真结果表明,相较于其它基准方法,本文提出的方案能够将优化目标值降低约7%,响应时延减少约12%,同时将缓存命中率提升约9%。
Vehicular Edge Computing(VEC)extends computing resources from the cloud to the network edge,such as Roadside Units(RSUs),providing nearby vehicles with computational support and enabling data processing and analysis at the edge.Consequently,offloading tasks to the network edge,such as roadside units(RSUs),for execution rather than processing them in the cloud or on vehicles themselves,can better meet the complex requirements of vehicular tasks.Particularly,this approach not only alleviates pressure on the backhaul network but also reduces the network latency.However,several objective factors still constrain the rapid development and performance enhancement of VEC systems.For instance,the computing power and storage resources at Edge Servers(ESs)are significantly lower than those in remote cloud centers.The mobility of vehicles results in a dynamic topology for vehicular ad-hoc networks(VANETs).The real-time requirements of vehicular tasks increase the complexity of allocating network bandwidth and computing resources at the network edge.To further enhance the performance of VEC systems,it is essential to consider the aforementioned factors and develop more efficient algorithms for task offloading and resource allocation in VEC.In this paper,to make full use of the limited resources of edge servers and improve the caching benefits of application services,a service caching based collaborative offloading model for vehicular edge computing was proposed.On this basis,the research on service caching and task offloading was conducted,with the purpose of minimizing the weighted sum of delay and energy consumption.Considering the reusability of application services,the application service can be cached at the edge server such as roadside units to serve the offloading requests.On the other hand,if the application service is not cached at the network edge,the vehicles can choose to offload the blocks of the required application services to the network edge in a collaborative way,on the assumption that the application service can be divided into smaller parts,i.e.,blocks.The proposed joint optimization of service caching,task offloading and computing resource allocation in vehicular edge computing was actually a nonlinear integer programming problem.To simplify this problem,the original optimization problem was divided into two subproblems.One is the joint optimization problem of service caching and task offloading in vehicular edge computing and the other is the optimization problem of computing resource allocation in vehicular edge computing.Specifically,the optimization problem of service caching and computing offloading was modelled as a Markov Decision Process(MDP),and a deep reinforcement learning-based caching and offloading algorithm was proposed.The optimization problem of computing resource allocation was proven to be convex,which can be solved by existing technologies such as interior point method.Extensive simulation was con-ducted to evaluate the efficiency and effectiveness of the proposed service caching and task offloading problem.Simulation results demonstrate that,in comparison with other baseline approaches,the proposed scheme exhibits superior performance,specifically by reducing the optimal values and response latency by 7%and 12%,respectively,and enhancing the cache hit ratio by 9%.
作者
唐朝刚
李召
肖硕
吴华明
TANG Chao-Gang;LI Zhao;XIAO Shuo;WU Hua-Ming(School of Computer Science and Technology,China University of Mining and Technology,Xuzhou,Jiangsu 221116;Mine Digitization Engineering Research Center of the Ministry of Education,Xuzhou,Jiangsu 221116;The Center for Applied Mathematics,Tianjin University,Tianjin 300072)
出处
《计算机学报》
北大核心
2025年第4期864-876,共13页
Chinese Journal of Computers
基金
国家自然科学基金面上项目(62476276,62271486)资助。
关键词
车载边缘计算
任务卸载
应用缓存
协作卸载
深度强化学习
vehicular edge computing
task offloading
service caching
collaborative offloading
deep reinforcement learning
