Research On Service Placement And Task Scheduling Strategy In Heterogeneous Environment Of Mobile Edge Computing

Mobile Edge Computer(MEC)places computing services closer to the user,allowing tasks that are offloaded by users to be processed locally.It is normally important to build some services on edge servers in advance in order to provide customers with low-latency,high-bandwidth,and high-performance computing services,so that computing services may be obtained faster when computing duties are offloaded to edge servers.At the same time,the execution efficiency of different service types under different computing architectures varies due to the diversity of service types.The computational resources available to distinct edge servers are diverse in order to improve overall execution efficiency.Edge servers,on the other hand,have restricted resources when compared to cloud computing.As a result,only a limited number of services are typically pre-deployed on edge servers.How to build a reasonable service placement strategy to satisfy the demands of users according to current user requests in the case of restricted and heterogeneous computing resources is a significant challenge that has to be solved promptly.Many computational tasks,on the other hand,can frequently be deconstructed into a set of subtasks with dependencies.Due to the previous task scheduling strategy's failure to take into account subtask dependencies,some subtasks will be scheduled to edge servers that are close to the user but have a long waiting list,resulting in a longer overall task completion delay.Furthermore,user movement will alter the type and quantity of job requests in various places.In hotspot locations,scheduling according to the principle of nearest priority will readily overload edge servers.As a result,determining the best job scheduling technique based on the present system state and reducing task completion delays is a problem that needs to be investigated further.This paper focuses on resource utilization in the online environment,taking into account the heterogeneity of the resources required by tasks in the mobile edge computing environment,the dependencies between sub-tasks,and the mobility of users,in order to improve the quality of experience(Qo E)of users.In-depth investigation on the deployment of services and job scheduling of restricted and heterogeneous edge servers,the main research contents are as follows:(1)This study provides a service placement technique based on reinforcement learning to reduce user-perceived latency in an environment where edge server resources are restricted and computing resources required for services are heterogeneous.First,the network,tasks,service placement,and processing power of edge servers are defined,and the current system information is modeled as a Markov decision process based on the heterogeneity of resources required by tasks,with the user-perceived delay as the assessment index.The reward and loss of deep reinforcement learning are designed as the computation delay required after service placement,and the service placement strategy with the highest reward is finally attained by neural network training.In comparison to previous benchmark methods,simulation results reveal that the suggested technique has a decreased userperceived delay.(2)This study provides a task scheduling method based on the Lyapunov optimization algorithm to reduce the user-perceived delay under the constraint that the subtasks have dependencies.To begin,the network is virtualized,after which the network model,task model,and computation model are created.Second,a work scheduling technique is proposed that is based on the Lyapunov optimization algorithm.The queue in the target problem is proposed for processing the Lyapunov drift plus penalty function.The state is limited in order to reduce the user's perceived delay while maintaining queue stability,and a theoretical analysis is performed.Finally,simulation experiments reveal that the proposed algorithm's edge server task waiting queue is substantially smaller than that of existing algorithms,while the total task completion rate is also improved.