| As a new form of network service management and deployment,network function virtualization(NFV)decouples network functions from dedicated hardware,provides a variety of services flexibly and efficiently for end users running resource intensive applications,and has been widely used in mobile edge computing(MEC)scenarios.By deploying multiple virtual network functions(VNFs)on specific edge servers,the effectiveness and flexibility of network services are greatly improved.According to users needs,multiple functional units are often called in sequence along a specific path,thus further forming a service function chain(SFC)to provide targeted network services,effectively reducing network service costs and improving scalability.However,the uncertain emergencies in practical applications will make the virtual network function fail and further cause the service function chain to be unable to realize its function smoothly,which seriously affects the users’ sense of service experience.Providing corresponding backup is an effective way to ensure the availability of network services.By deploying virtual network functions and service function chains backup on the edge network,it can ensure that corresponding functions can recover quickly from unexpected failures and provide users with continuous and reliable services.However,due to the limited resource constraints of edge networks,it is inevitable that not all backups can be implemented.Therefore,it is crucial to choose the best set of virtual network function units for backup,For selected multiple backups.how to properly deploy them on specific edge servers to provide users with the most efficient services in a chain structure is also a problem that cannot be ignored.The existing researches have largely ignored the impact of different requirements for virtual network functions and service chains and the uncertain and unpredictable failure conditions on the backup selection problem.At the same time,the workload problem in the edge environment and the latency that directly affects the service efficiency are of ten not properly solved.This thesis studies the implementation of network function virtualization in edge environments,focuses on the selection and deployment optimization of backups,proposes a new concept called demand level as the consideration of the importance of different virtual network functions and service function chains by comprehensively considering both the service needs of end users and the combined needs of chain structures,and addresses corresponding challenges from the perspectives of functional units and functional chains based on the combinatorial multi-armed bandit learning model.To start with,for independent virtual network functions,this thesis formulates an online learning-based virtual network function backup selection and deployment algorithm,which places the optimal set of units on specific edge servers properly and greatly improves resource utilization.At the same time,considering the low efficiency caused by the uneven workload between different edge servers,this thesis extends the above problem and proposes corresponding extended algorithm.Subsequently,for the service function chains formed by multiple virtual network functions,we further consider real-time failures and latency consumption and formulate a real-time selection and deployment algorithm combining the greedy algorithm and the heuristic Prim algorithm.It provides an approximately optimal solution for the backup problem at the edge of functional chains,thus offering end users popular services with the lowest latency.A large number of simulation experiments have proved the feasibility and effectiveness of the proposed algorithms. |
