Research On Scheduling Optimization With Edge-Cloud Collaboration

Cloud computing is a computing paradigm based on network,which centralizes computing resources,storage resources and network resources to the cloud data cen-ter for unified processing.To some extent,cloud computing solves the problems of insu cient device memory,weak computing power and low e ciency.What's more,as the number of devices grows,the delay is also increasing,so that users'requests can't be satisfied in time.Edge server is an e ective way to bridge cloud and users.Through edge computing,some workloads can be o oaded from the cloud to the edge,so as to reduce the pressure on the cloud center and save bandwidth,enhance the user satisfaction.This is very beneficial to delay sensitive applications such as intelligent transportation and intelligent safety analysis.Edge computing and cloud computing are not competitive with each other,but mutually supportive.Compared with cloud servers,edge servers have limited computing resources and insu cient network bandwidth.When the users in the system put forward the demand task,whether the computing task is executed locally or o oaded to the edge server or cloud,the mobile device should be planned,which is the core problem in edge computing-task o oading.In addition,after the task arrives at the edge server,the execution of the task needs data,and the memory resources of the edge servers are limited,so the long-term storage of data will lead to a large storage cost.How to use the limited resources to cache data at the edge of the network to complete the real-time requests of surrounding users is also a key research direction,which is called caching problem.Task o oading and data caching are two kinds of very basic optimization problems in edge-cloud computing scheduling,which are closely related to resource allocation.Firstly,we extend the traditional o oading model to a more general heterogeneous o oading model.In traditional studies,the communication cost between any two tasks is often assumed to be symmetric in di erent sides and is often ignored within the same side.In this paper,we generalize the traditional model,and assume that the communication cost exists everywhere and is asymmetric.With the help of semidefinite relaxation,we give an algorithm for the o oading problem.If the Laplacian matrix with respect to the o oading problem is positive semidefinite,the approximation ratio of the algorithm is(?).Secondly,we study data caching problem by extending single data item to multiple data item among servers.About the homogeneous model and the submodular model with constraint,we propose a data caching strategy to minimize the total transfer and caching costs of the system.Moreover,the semi-heterogeneous model can be solved by the anticipatory caching(AC)algorithm in our previous work.Meanwhile we find it is more e cient for our three models in this thesis to improve the performance.Finally,based on the data caching problem with single data item,we consider adding the data center in the cloud system.The edge node can provide certain accessible storage and computing resources for the surrounding users.However,the resources of edge servers are limited and sometimes have to gain the data by transferring from other servers or buying from data centers at the cloud.Traditionally,data center is just regard as a place of purchasing data,and the data is always available for the buyer after purchase.However,in many practical application scenarios,the data is often subject to deadline constraints,or,after the buyer purchases the data,it only provides limited calls for free.Based on this assumption,in this thesis,we study the data caching problem in the edge-cloud collaborative system to minimize the total cost.Without knowing the information of users'future request flow,we propose an online algorithm.Through rigorous theoretical analysis,we prove that the asymptotic competitive ratio of the algorithm in the worst case is 3.