Computing Offloading Of Composite Technology In Edge Computing System

As the wide deployment of the Internet of Things and the promotion of 5G mobile communications services,the number of devices attached to the Internet increases explosively.For the processing of massive amount of data generated in the Internet of Things,it is usually difficult for end devices to undertake these job due to their limited computing resources.Meanwhile,limited by long network distances between data centers and Internet edge,it is challenging for cloud computing to meet the requirements of delaysensitive applications,and associated data transmissions heavily load the Internet.Being a new pattern of computing,edge computing brings computing resources network edge,processes massive amount of data produced by end devices nearby,and thereby matches diverse requirements of Internet of Things applications.Taking into account available computing resources and network locations,computing offloading determines how to assign computing tasks among devices,edge nodes and cloud,and consequently become underpinning part edge computing.Therefore,edge computing offloading has drawn wide attentions in both academia and industry.Due to the diversity of Internet of things application computing processing,the computing tasks processed by edge computing system present a variety of different modes,including the composite mode composed of multiple subtasks with dependency.This thesis focuses on the offloading technology of two kinds of complex computing tasks in edge computing system,including overlapped sequential and multiple input multiple output universal.(1)Overlapped sequential computing tasks are composed of multiple overlapped(partially parallel)subtasks with sequential dependencies.In this thesis,we model the offloading problem of overlapping sequential composite computing tasks in edge computing,and consider the influence of subtask requirements,edge node basic load and network transmission.Taking the completion time of computing tasks as the optimization objective,we propose a dynamic programming based task offloading algorithm for overlapping sequential computing(OSTODP),and carry out simulation experiments.The simulation results show that OSTODP can significantly shorten the completion time of overlapped sequential composite computing tasks under different subtask overlapping ratio,computation load and basic load distribution of edge nodes.The proportion of overlapping subtasks,the basic load distribution between edge nodes,and the overall system load all have significant effects on the completion time of overlapping sequential computing tasks.(2)Multiple input multiple output general computing tasks are composed of multiple subtasks whose dependency relationship does not exist a specific pattern and whose execution order can be represented by a general directed acyclic graph(DAG).In this thesis,the multiple input multiple output composite general purpose computing task offloading is expressed as a mixed integer programming problem,A computational task offloading algorithm(SM-BBM)based on simplex method and branch delimitation method is proposed to solve the subtask execution node allocation scheme.The corresponding simulation results show that SM-BBM can significantly shorten the overall completion time of multiple input multiple output composite general computing tasks under different computational load,existing system load and the distribution of computational load among subtasks.Compared with the branch bound algorithm(BBM),SM-BBM can significantly reduce the number of branches and the running time of multiple input multiple output composite general computing task offloading.In the two kinds of composite computing tasks considered,the overlapped sequential type actually covers two common types: parallel type and pure sequential type.The multiple input multiple outputs universal type can abstract many data processing computing tasks in the fields of smart city and industrial Internet.Therefore,the proposed algorithm OSTODP and SM-BBM can be effectively applied in the actual edge computing system to optimize the service performance and resource utilization.