随着边缘计算的广泛应用,近年来在网络边缘侧激增了一些延迟敏感的用户请求,这些应用对边缘网络中物联网设备提供的资源提出了较高的服务质量(Quality of Service,QoS)需求,例如严格的地理空间约束、时延/能量及其他资源约束.物联网设备提供的功能通常被封装为运行在边缘节点上的服务,用户请求可以通过组合数据和/或计算密集型的物联网服务来实现.考虑到物联网设备的资源稀缺性以及用户请求的在线持续部署和潜在长期执行特征,边缘服务运行期间对物联网设备资源的占用和释放导致边缘网络中资源动态变化.由于物联网设备的资源通常难以得到有效补充,且消耗差异可能较大,有些设备可能会过载,导致在当前时间点适配的物联网服务,在随后时间点可能难以适配用户请求,并导致QoS降级.针对边缘网络高负载时新请求持续部署导致特定强约束难以满足的挑战,本文开展资源失配时低代价的服务重配研究,提出了一种资源高效的服务重配方法,旨在通过服务迁移技术重调度物联网设备所提供的服务,以满足更多具有一定QoS约束的用户请求.基于上海电信基站数据集进行了大量实验,实例验证本文方法的有效性.实验结果表明,本文所提方法在满足用户服务请求时延约束、降低物联网设备能量消耗、提高边缘网络资源利用效益等方面表现均优于对比技术.
Data and Computation-Intensive Service Reconfiguration with Low Cost of Imbalanced Resources in Edge Networks
The advent of edge computing has revolutionized the landscape of Internet of Things(IoT)applications,enabling a plethora of services to operate at the network periphery.Among these are augmented reality,online interactive gaming,and real-time video processing,all of which are characterized by their sensitivity to latency.The quality of service(QoS)demands for these applications are stringent,necessitating precise control over geospatial positioning,re-sponse times,energy efficiency,and other resource-intensive constraints.IoT devices,which are integral to the functioning of these applications,encapsulate a variety of functionalities through IoT services.The fulfillment of user requests often hinges on the seamless composition of these data and computation-intensive services.However,the resource limitations inherent to IoT de-vices pose a significant challenge.The dynamic allocation and deallocation of resources during the runtime of IoT services lead to fluctuations in the availability of these resources within edge net-works.Given the difficulty in replenishing IoT device resources and the potential for substantial variability in their consumption,the risk of device overload is a genuine concern.This can result in a decline in the ability of IoT services to consistently meet user requests,thereby degrading the QoS for both current and future requests.This paper introduces a novel low-cost service reconfig-uration strategy designed to mitigate the imbalance of resources in edge computing environments.By leveraging service migration technologies,we propose a resource-efficient reconfiguration method capable of accommodating a greater number of forthcoming user requests,all while adhe-ring to specified QoS constraints.We formulate the service reconfiguration problem as markov multi-phases decisions,which are addressed by Double Q-network with the replay buffer to en-hance Reinforcement Learning(denoted DQRL).This algorithm considers the delay of IoT appli-cations and resource utilization of IoT devices comprehensively,to optimize service reconfiguration in edge networks.The proposed strategy is underpinned by an intelligent deci-sion-making framework that optimizes the allocation of resources to IoT services,thereby enhan-cing the resilience and adaptability of edge networks to fluctuating demands.To validate the effi-cacy of our approach,we conducted extensive experiments using a dataset from Shanghai telecom base stations,which provided a realistic and complex environment for testing.The experimental results show that our approach performs better than baseline techniques in terms of satisfying de-lay constraints of IoT applications,decreasing energy consumption,and improving the resource utilization efficiency of IoT devices.This performance not only contributes to the sustainability of IoT ecosystems but also enhances the overall resource utilization efficiency of IoT devices.
data and computation-intensive serviceresource utilization efficiencyservice recon-figurationservice migrationedge networks
万方数据
维普
