Research Of Cloudlet Placement Strategy Based On Spectral Clustering In Mobile Edge Computing

In recent years,driven by the Internet of Things and 5G communication,the paradigm of mobile cloud computing has changed from centralized mobile cloud computing to distributed mobile edge computing.Mobile edge computing is an emerging architecture that migrates computing,network control,and storage to the edge of the network and provides a low-latency,high-bandwidth cloud computing environment for resource-constrained mobile devices within the access network.Although mobile devices can offload tasks to remote cloud execution to save energy on mobile devices,the migration of user tasks to the central cloud will result in greater transmission delays.As a new component of mobile edge computing,cloudlet can perform the task of unloading mobile users to reduce the access delay of mobile users and meet their requirements for system response time.At present,most researches on mobile edge computing focus on how to segment and offload computationally intensive applications to resource-rich cloud platforms for the purpose of reducing the energy consumption of mobile devices and reducing the completion time of mobile applications.However,how to deploy cloudlet in a large-scale network and improve the quality of mobile application services is still a difficult point in current research.Aiming at the problem of cloudlet placement in mobile edge computing,this paper designs a multi-cloud multi-user placement model and corresponding placement algorithm to achieve normalized cut value and average access delay minimized;this paper also designs a multi-cloud multi-user cloudlet load balancing algorithm to minimize system response time.Its main research contents and contributions are as follows:(1)Aiming at the network scale problem of wireless metropolitan area network,a cloudlet placement model based on spectral clustering is designed.The matrix perturbation theory is combined with the location attribute of the access point,and the localized segmentation and cloudlet placement position are obtained by minimizing the normalized cut value and minimizing the sub-area user access delay as optimization targets.This model fully considers the physical distance between each access point and the shortest data transmission delay,and uses the matrix property to solve the network expansion.(2)Aiming at the cloudlet placement model and optimization problem of wireless metropolitan area network,a cloudlet placement algorithm based on spectral clustering is designed.The algorithm considers the influence of the number of access points,the connection status between access points and the user's request arrival rate of the access point to optimize the access delay of the mobile user's offloading task to the cloudlet.The access point deploys a cloudlet.Because the proposed algorithm has low time complexity,it has a good application prospect for the cloudlet deployment of large-scale wireless metropolitan area networks.(3)Aiming at the problem of unbalanced cloudlet load in wireless metropolitan area network,a cloudlet load balancing model and corresponding load balancing algorithm are constructed.Based on the proposed cloudlet placement model,considering the user task arrival rate of each cloudlet,the user unloading tasks of some cloudlets are reallocated to achieve load balancing of the cloudlet.The algorithm significantly reduces task completion time and reduces energy consumption of mobile devices.