Design And Implementation Of Mobile Edge Computing Models With Self-adaptive Task Offloading

With the rapid development of the Internet of Things and communication technologies,mobile devices,such as smart phone,are becoming more and more powerful.People are increasingly demanding them to handle complex computing tasks.Therefore,cloud computing has been applied to all aspects of mobile services due to its high computing power,low service cost and high availability.However,in recent years,the demand for rate and quality of mobile service has continued to grow exponentially.Cloud computing services are gradually unable to meet the user requirements due to high jitter,high delay and high energy consumption caused by long-distance transmission.The newly proposed Mobile Edge Computing,on the basis of traditional cloud computing,performs data processing through the network edge near data source,to make the cloud resources and services close to the end users.In this way,it can effectively shorten the task execution delay and energy consumption of the mobile device.Mobile edge computing leverages energy-constrained mobile devices by offloading computationally intensive tasks to nearby resource-rich edge servers.However,as a new research area,its infrastructure has not yet been fully established.Researchers must investigate the task offloading problem by modeling the edge computing scenarios at first.Most of the existing offloading mechanisms only proposed the time model and energy consumption model.The whole process of the task offloading and the scalable system implementation of each model have not been studied to simulate the reality.Systems generated in cloud computing and fog computing cannot be applied to mobile edge computing due to different scenarios.In addition,the existing offloading algorithms rarely adjust the offloading scheme and select the optimal offloading mechanism according to the past situation.In this paper,we analyze the mobile edge computing scenario,formalize the task offloading problem meeting resource constraints and establish a mobile edge computing architecture for offloading,including mobile device,transmission network,edge server,load balancer and remote cloud.The mobile device generates all tasks and data.The transmission network records the network topology and calculates delay for uploading and downloading data.The edge server provides the computing service near data source.The load balancer coordinates the virtual machine load in the edge server to maintain the overall performance.The remote cloud is a traditional cloud computing platform far away from the mobile device.What's more,we design models and policies in mobile edge computing,including mobility model,task generator policy,network model,virtual machine allocation policy,time model and energy consumption model.The mobility model simulates the movement process of mobile device.The task generator policy segments all tasks from mobile device.The network model simulates the network transmission process and calculates delay.The virtual machine allocation policy is to allocate or reallocate virtual machines in the host of edge server.The time model and energy consumption model calculate time and energy consumption of mobile device for the task offloading process.To make up for the lack of system in mobile edge computing,we design a relatively complete system called MECSim,including all designed models and policies.To solve the offloading problem,we propose the Worst Fit,Best Fit and First Fit task offloading algorithms in mobile edge computing based on the dynamic memory allocation algorithms,to select the appropriate virtual machine for the offloadable tasks.In order to deal with more complex scenarios,we propose the self-adaptive task offloading mechanism according to the power of mobile device,the failure rate of recent tasks and the size of task,combining the Worst Fit,Best Fit and First Fit task offloading algorithms.This mechanism automatically adjusts the offloading way to adapt to different scenarios,saving time and energy.Finally,we analyze the overall implementation,module partition and key modules of the MECSim system with the factory pattern,including related models,policies and offloading schemes,from the perspectives of class diagram,class analysis and sequence diagram.Experiments are carried out on MECSim to verify the importance of mobile edge computing,the effectiveness of the system architecture,the dynamic nature of the model and policy,and the efficiency of the self-adaptive task offloading mechanism proposed in this paper.