Research On Resource Allocation Of Network Function Virtualization Based On Prediction

Network Function Virtualization(NFV)uses virtualization technology to provide a new way for the design,deployment,and management of network services.It separates the special physical network devices and the network functions running on them,and makes these network functions(such as poxys)deployed on the general physical server in the form of software.When users need multiple network function services,these network functions can form a network function service chain.NFV technology can instantiate net-work functions(such as firewalls,proxys,etc.)on the chain to servers so that these servers can be conveniently placed in the data center and distributed network nodes.Therefore,NFV technology can also achieve the goal of reducing Operating Expense(OPEX)and Capital Expense(CAPEX)without purchasing new hardware and promote the agile and efficient deployment of network functions.Due to the advantages of NFV,many Network Function Virtualization Provider(NFVP),deploy Virtual Network Function(VNF)on the cloud by purchasing cloud resources to provide NFV services,to save the cost of deployment.At present,most of the models about the resource allocation of NFV are offline,while the research on the dynamic system is less.Therefore,how to design an algorithm to deal with the resource allocation under the virtualization of network function in a dynamic system is very challenging.Aiming at the dynamic system,this thesis studies the resource allocation problem in the scenario of user location dynamic change,and the resource allocation problem in the scenario of user request dynamic fluctuation and cloud server price dynamic change.The main work and innovation are as follows:First,aiming at the scenario of network slice resource allocation in the system with dynamic change of user location,a solution strategy of prediction combined with model predictive control algorithm is proposed.By considering the constraints of computing resources and delay,we build a mathematical model to minimize the system cost to deploy VNF.First,we use Long Short Term Memory(LSTM)to predict user mobility.Then,we design an Model Predictive Control(MPC)algorithm based on prediction information to solve the proposed mathematical model.Finally,we verify the effectiveness of the algorithm through simulation experiments to achieve the purpose of saving system cost.Second,for the scenario that NFVP deploys VNF to provide NFV services on the cloud by purchasing cloud resources,this thesis considers the dynamic characteristics of cloud resource prices and user requests.We establish a mathematical model aiming at minimizing the cost of purchasing cloud resources.Firstly,we use LSTM to predict the price trend of cloud resources and the change of user traffic requests.Secondly,we pro-pose a deep reinforcement learning strategy based on predictive information to reduce the service cost.Finally,through a large number of experiments and simulations,we verify the performance of the algorithm and achieve the purpose of reducing the service cost.