Quantitative Analysis Of Cloud-Edge Service Performance Based On Stochastic Model

With the rapid development of mobile communication technology,devices connected to the Internet(such as smart vehicles,smart phones,intelligent traffic lights,etc.)are growing explosively,and they have stronger real-time requirements.Traditionally,the cloud has been the preferred solution for providing storage and computing resources to process the massive amounts of data generated by Io T devices and perform the necessary analysis.However,offloading the tasks generated by these devices to a remote cloud for processing suffers from high communication latency and energy consumption,which violates the latency requirements of real-time Io T devices.To cope with the stringent latency requirements of devices,edge computing has been proposed,which offloads some computing tasks of devices to the edge computing server for service.Relying on the mutual cooperation between cloud computing and edge computing(cloud-edge architecture),Quality of Service /Quality of Experience can be guaranteed for user.However,as cloud-edge architecture plays an increasingly important role in industry transformation and upgrading,quantitative analysis of cloud-edge service performance becomes more and more urgent.This thesis aims to use methods such as stochastic modeling and simulation experiments to quantitatively evaluate the performance of cloud-edge services in terms of task rejection probability and mean response time.The main research contents and contributions include the following three aspects:(1)A quantitative analysis method is proposed that can describe the impact of different cloud-edge computing resource allocation strategies(sequential selection strategy and random selection strategy)on the performance of tasks with time-varying and real-time arrival.In the cloud-edge system,the time interval of task arrival changes with time and there is a tolerance time in the system.In addition,according to the processing capacity of the edge data center and cloud data center,the decision-making node decides that the task enters the edge or cloud data center and selects the physical devices in the data center sequentially or randomly to provide services for the task.When quantifying and evaluating task service performance,existing models either ignore cloudedge resource allocation policies,ignore task arrival time variability,or ignore the tolerance time when tasks are processed in the system.Firstly,this work analyzes the behaviors of tasks arriving,waiting for processing,deployment,service and waiting time exceeding the maximum deadline in each data center under this scenario,and constructs sub-models for cloud data center,edge data center and decision node respectively;Then,the calculation formula of key performance indicators is designed based on the stochastic model.The performance indicators include task rejection rate and average response time;Finally,the approximate accuracy of the hierarchical model and performance metric formula is verified by comparing the simulation experimental results and numerical experimental results,and quantitatively evaluate the impact of different system parameters on the performance of cloud edge services.(2)A quantitative performance analysis method for tasks with priority,heterogeneous number of resources and real-time is proposed.In the cloud-edge system,the number and type of physical or virtual resources requested by tasks may be different.In addition,when evaluating task performance quantitatively,existing models either ignore the priority of tasks,or ignore that tasks require different numbers of resources when serving,or ignore that there is a tolerance time when tasks are processed in the system.On the basis of Contribution 1,firstly,this work analyzes the processing order of tasks under the principle of first come first serve and non preemptive priority,and analyzes whether tasks can be served according to the available resources of physical devices.Then,a hierarchical model that can capture the behavior of tasks entering/processing/leaving decision nodes,edge data centers and cloud data centers is constructed,and the formulas of key performance indicators are deduced.Finally,based on numerical analysis and discrete event simulation experiments,the approximate accuracy of the proposed hierarchical model and performance index calculation formula is verified,and quantitatively evaluate the performance of cloud-edge service under different system parameters.(3)A quantitative performance analysis method for real-time tasks with mobility,priority and different number of resources is proposed.In the cloud edge system,in order to solve the problem that user mobility may lead to service quality degradation or service interruption,multiple mobile edge data centers are required to cooperate with each other to realize dynamic service migration.Existing models only analyze that task is processed until completion in the same mobile edge data center,and cannot be used to capture the migration behavior of tasks in multiple mobile edge data centers;in addition,synergistic effects of the mobility,real-time,priority and resource number heterogeneity on task performance is ignored.On the basis of Contribution 1 and Contribution 2,firstly,this work analyzes the decision behaviors of different decision nodes and the migration behavior of tasks among edge data centers.Then,we build an iterative model between edge data centers,which can capture the data interaction process between edge data centers,and then build an extensible hierarchical model that can capture the arrival,processing and migration of tasks in the cloud side system.On this basis,the calculation formula of cloud side service performance index is deduced and proved theoretically.Finally,numerical experiments and simulation experiments are used to verify the approximate accuracy of the formula and model,and further quantitatively evaluate the impact of the number of mobile edge data centers and other parameters on the performance of cloud edge services.