Resource Management Based On User Mobility Characteristics In Mobile Edge Computing

With the rapid development of Internet of Things(IoT)technologies,the number of mobile devices such as smartphones,tablet computers,and IoT terminals continues to grow.Meanwhile,new applications such as computing-intensive and delay-sensitive applications emerge in large numbers,e.g.,virtual reality,augmented reality,online games,and Internet of vehicles applications,etc.Network is required to provide low-latency,high-reliability services.The traditional mobile cloud computing is built on the central cloud far away from the mobile users,which leads to excessive delay and energy consumption because of the data transmission through the wide area network,thereby affecting the user’s application experience quality.In this context,the concept of Mobile Edge Computing(MEC)was proposed.MEC deploys cloud computing capabilities to the edge of networks,shortenings the physical distance between mobile devices and cloud services,and thus to meet the requirements for low latency and low energy consumption of mobile devices and applications.MEC has become an indispensable supporting technology for the IoT,Fifth Generation Mobile Communication Technology(5G),Industrial Internet of Things(IIoT),and artificial intelligence.The main resource management technologies in MEC include service deployment,computation offloading,service migration,etc.Edge service deployment,which deploys applications on the appropriate edge cloud servers.However,the dynamic spatiotemporal changes of user service requests make it difficult to tradeoff the service deployment cost and service efficiency.Computation offloading,which offloads the tasks on mobile device itself to other intelligent nodes such as the edge cloud servers or mobile devices to be processed.However,the limited computing resources and bandwidth resources in edge networks make it impossible for a single node to efficiently complete all the tasks from mobile users.Service migration,which migrates the running edge service from the source site to the target site.However,the passive migration method leads to long service delay,which is difficult to meet the requirements of delaysensitive applications.How to solve the above problems is a severe challenge to realize efficient utilization of MEC resources and enhance user services quality.User mobility is the essential attribute of the mobile networks.Especially in edge computing,it is the user mobility that causes the dynamic changes in resource requirements of mobile devices.And the studies have found that user mobility present certain spatiotemporal characteristics.For example,people tend to use the certain application at different times and different locations,the similarity of user mobility has strong implications for social relationship,the mobility is predictable,etc.The above mobility characteristics provide an important basis for edge computing resources to be configured on demand.Therefore,this paper aims to solve the resource management problem in MEC from the perspective of user mobility characteristics.In this paper,research on MEC is from three aspects:service placement,computation offloading,and service migration.The main contents are as follows:1.Aiming at the problem that it is difficult to balance the cost and efficiency of edge cloud service deployment caused by the dynamic spatiotemporal changes of user service requests,this paper proposes an edge service placement method based on the spatiotemporal characteristics of user mobility.Introducing the regular characteristics of the impact of user mobility on application usage behavior,the delay model and deployment cost model are established.A service placement method based on multi-object Context Multi-armed Bandit learning with a Dominant Objective(CMBDO)is proposed.More specifically,taking the mobility information as the context information,the delay as the dominant objective and the deployment cost as the non-dominant objective,a local optimization method based on multi-objective learning is designed.In this method,the mobility information is divided into blocks with similar mobility,and the requirements in each block are learned by observing the benefits obtained from the past decision history and the current service requests.Then the ideal edge servers are selected to place application services.Finally,the convergence of the proposed method is proved by theoretical analysis,and the simulation results show that the proposed method is significantly better than the benchmark methods in terms of service delay and deployment cost.2.Aiming at the problem that mobile cloudlet fails to process the offloaded task due to the multiple constraint conditions,the offloading availability assessment based on mobility similarity is proposed.Moreover,the formal models for minimizing energy consumption and traffic of mobile devices are established.Especially,the multi-site offloading problem is formulated as a multi-objective optimization problem.A lowcomplexity task offloading algorithm based on multi-objective simulated annealing is proposed,meanwhile the pareto solution set is obtained.More specifically,the values of three objective functions are obtained from the randomly generated initial solution vector,and new objective function values are calculated according to the new vector generated by the current optimal solution.Then the annealing plan is utilized to determine whether to accept the new solution.At different temperatures corresponding to each objective function,iterations with a certain length are performed.According to the annealing schedule,the temperatures corresponding to the three objective functions are respectively lowered until the temperature is lower than the given threshold value.Finally,the local optimization scheme of task offloading is obtained,and the simulation results verify that the proposed algorithm has low complexity,which outperforms the benchmark algorithms in terms of offload availability,energy consumption,and traffic.3.Aiming at the problem that the long service delay caused by the passive service migration method is difficult to meet the requirements of delay-sensitive applications,a service migration method based on mobility prediction is proposed.The minimization models for end-to-end delay and service migration delay are established,and the target edge server based on mobility prediction is selected,and the optimization for container migration procedure are jointly considered.Specifically,an ensemble learning algorithm combining a recurrent neural network with a preference-embedded Markov chain is utilized to obtain the target edge server.Then the service migration procedure is optimized by migrating the basic container image and application image in advance of communication handover,and the timing planning scheme in service migration procedure is proposed.By checkpoint restore technology in container,migration algorithms in the source server,the target server and the controller are designed,thus realizing the proactive service migration.Finally,the simulation results verify the effectiveness of the proposed scheme in providing low latency.