Research On Scheduling Of Correlative Tasks In Multi-access Edge Computing

Currently,the widespread use of Io T and smart mobile devices has greatly facilitated the development of many latency-sensitive and resource-intensive applications,such as virtual/augmented reality,face recognition,etc.As mobile applications become more and more complex,the required resources are also increasing.However,the smart mobile devices are usually resource-constrained,such as processing speed,memory size,and battery power,etc.,which make the smart mobile devices lack sufficient computing resources for executing complex applications.Edge Computing(EC)has become a new computing paradigm for executing delay-sensitive and resource-intensive applications.However,the tasks of many applications can be divided into multiple correlative tasks,for example,face recognition application can be divided into five correlative tasks:object acquisition,face detection,preprocessing,feature extraction and classification.When these tasks are offloaded to edge nodes,the dependence relationship between tasks need to be considered.For example,the output of task “feature extraction” is the input of task “classification”.Thus,the task“classification” can start only if the task “feature extraction” has completed.Most of the previous research work ignores the effect of task correlation and edge node heterogeneity on offloading strategies,which may lead to infeasible scheduling decisions.This thesis mainly studies the scheduling decisions of correlative tasks in Multi-Access Edge Computing system composed of multiple terminal devices and heterogeneous edge nodes.Firstly,this thesis studies the scheduling of correlative tasks on heterogeneous edge nodes,formulates the problem as Correlative Task Scheduling in HEterogeneous environment(CTSHE),and show that this problem is NP-hard.Then,the Speed-based Earliest Time First Scheduling(SETFS)algorithm is proposed to solve the CTSHE problem,which provides an approximate guarantee of the algorithm in the worst case.Finally,the experimental results of the SETFS algorithm in makespan and running time are analyzed through simulation experiments.The simulation results show that SETFS can significantly improve the makespan of tasks.Moreover,SETFS is suitable for larger-scale correlative task scheduling with high heterogeneity of processing speed and unit transmission time.Then,this thesis studies a more complex system model.Since some tasks on the application must be finished locally,the joint scheduling method of correlated tasks on the local and edge sides is studied,and studies another energy consumption problem that can not be ignored in the offloading scenario of multi-access edge computing.This thesis formulates the problem as minimizing the energy consumption in the system under the condition of satisfying the deadline constraints of the application,the Joint Scheduling Algorithm(JS)is proposed to solve this problem.Finally,the experimental results of the JS algorithm in the application finish rate and system energy consumption are analyzed through simulation experiments.The simulation results show that the JS algorithm is superior to other comparison algorithms in the application finish rate and can save at least 43% of the system energy consumption.