| With the rapid development of advanced communication technologies and the explosive growth of innovative mobile applications,edge computing is leading the way towards more powerful,lighter,and better experiences for mobile applications.It takes computing offload technology as a breakthrough point to bring low-latency local computing power,storage,and network resources to mobile devices and completely changes the way of designing and developing applications for end devices.Energized by the novel computation offloading technologies,edge computing has introduced high-response and low-latency local network,computing,and storage resources to mobile applications,breaking the situation that the end devices can only be powered by limited local resources.As the key technology of edge computing,edge computation offloading aims to accelerate mobile applications and improve user experience by offloading computation-intensive tasks to loacal computing facilities.There are two implementations of edge computing offloading,namely direct offloading of computing tasks and the offloading of computing tasks in the form of service invocation.However,due to the limited edge resources and timely demands of offloading results,how to realize the optimal scheduling of task offloading and the placement of mobile services become the key challenges.With the continuous improvement of edge computing specifications and the continuous implementation of related technologies,more and more applications will support the edge acceleration function.Therefore,more and more traditional cloud computing providers have begun to extend their business and infrastructure to the edge,which makes task offload scheduling and service placement technology in the edge computing environment hot research issues.This thesis focuses on the task offload scheduling and service placement at the edge computing offload environment and studies its scheduling algorithm,especially focusing on the decentralized online methods in the edge environment with high dynamics.The main contributions of this thesis include the following four aspects:1)For the online mobile task offloading problem,we propose a decentralized algorithm based on dynamic learning.Considering the limitation that most of the current scheduling algorithms adopt a centralized architecture and suffer from single point of failures,the algorithm proposed reckon that the task/resource state information is usually effective locally,and employs a decentralized architecture to yield online schedules.At the same time,considering the dynamic and time-varying characteristics of the supply and demand of edge resources,a dynamic learning-based method is proposed to update the scheduling strategy at runtime.Thus,it is capable of making a quick response to rapid edge environmental changes.Experiments have shown that the proposed online scheduling algorithm with decentralized architecture achieves higher scheduling responsiveness and lower average unloading delay than baseline ones;2)Aiming at the multi-tenant computing offloading task scheduling problem constrained by different service level agreements,a decentralized method consisting of three strategies,i.e.,online task scheduling/migration and resource allocation,is proposed.Considering that most traditional computational offload task scheduling algorithms based on the assumption that the properties or distribution of offload tasks are known in advance,and the expected execution time of the task can be accurately estimated.To eliminate the limitations caused by these assumptions,all three strategies of the proposed method are independent of the task execution time.At the same time,the proposed method also fully considers the heterogeneous resource deployment situation in edge scenarios and the diversity of user offloading tasks,which supports different tasks have different acceleration effects on different architectures of edge resources.Experiments show that the proposed method considering multi-tenancy and heterogeneous resource environments can achieve lower user priority weighted task offloading response time under various load conditions with controllable task migration cost;3)Aiming at the online edge service placement problem,a user mobility perception method based on edge-user collaboration is proposed.This method consists of an edge-side coordination algorithm and a device-side mobility-aware service real-time placement decision algorithm.Considering end users’ requirement for low response delay of edge services,and the distance constraints between edge services and end-users,a decentralized scheduling architecture is designed to delegate the decision-making power of service placement to user-end devices,while the service migration coordination strategy is remained at the edge side.At the same time,considering the ubiquitous mobility of edge users and its huge impact on the placement results,this method also takes the user’s instantaneous movement characteristics into consideration when making service placement decisions.Experiments show that,compared with the algorithm that does not consider user mobility,the proposed algorithm achieves a higher service placement success rate and lower service placement times;4)For the problem of energy and cost-aware online edge service placement,we propose a fuzzy control-based,and decentralized online service placement method.Similarly,this method does not have a centralized scheduler and state sharing mechanism either,and it can be deployed on user-devices in the form of plug-ins of mobile applications.When users need to invoke edge services,they can make their own real-time placement decisions locally by the states of nearby available edge servers.Considering the dynamically changed edge resource environments,the proposed method contains two heuristic placement strategies,in which the service consolidation strategy is suitable for saving costs and reducing energy consumption when the load is low.While the load balancing strategy is employed to guarantee the guarantee when the request load is relatively high.Besides,a fuzzy control method is proposed to realize the step-less adjustment of the weights of these two strategies.Experiments based on real-world edge environment datasets have verified the the effectiveness and efficiency of the proposed method. |
