Research On Virtual Network Function Deployment And Consolidation Optimization Method

Network Function Virtualization(NFV)leverages Virtualization technology to deploy network functions on standard commercial servers,forming virtualized network functions(VNFs)to replace the dedicated middlebox devices.Multiple virtual network functions can coexist in a common server in the form of virtual machines,and can be installed or moved to any location in the network as needed without the need to deploy new hardware devices,thereby reducing equipment investment and operating costs for the enterprise.Usually,traffic is processed by network functions in a specified order to enhance application security and performance.And an ordered set of network functions form a service function chain(SFC).In NFV-enabled network,SFC is in the form of an ordered set of VNFs.Given a series of traffic service requests containing SFC requests,we should deploy related VNFs and traffic paths to serve them.The software feature of VNF make it more flexible to decide where to deploy it.Different deployment locations can cause different energy and resource consumption.Therefore,for the traffic service requests containing SFC requests,an effective method to determine the deployment location of VNF is required to reduce energy consumption and resource consumption,which is refered to i ITNTial VNF deployment optimization problem(or VNF deployment optimization problem for short).In addition,in the long-term running of the network,the load of servers having VNF deployed changes dynamically with the change of communication traffic,making the i ITNTial VNF deployment scheme no longer efficient.When there is a large number of server load reduction,it is necessary to integrate VNF into fewer servers to shut down some servers,thus reducing the use of servers and reducing the energy consumption of servers.This process is called VNF consolidation.Choosing which servers to shut down and how to migrate the VNF directly affects the VNF migration costs and energy savings involved in VNF consolidation.Therefore,an optimal VNF consolidation solution is needed to maximize energy savings and minimize the migration costs.This problem is also known as VNF consolidation optimization problem.Currently,the deployment optimization and consolidation optimization of VNF in NFV environment still have the following problems:(1)For effectively balancing the server energy consumption and bandwidth resource consumption,the existing VNF deployment optimization method still has space to be improved.(2)Due to the NP-hard characteristics of the VNF deployment optimization problem and the huge solution searching space,the time efficiency of the existing meta-heuristic or heuristic VNF deployment optimization method is still not high.(3)In datacenter,for the deployment optimization of VNF required by the traffice between Virtual machines(VM)carrying application business,the existing work ignored the deployment location of the VM carrying application business has influence on the VNF deployment;(4)The existing VNF consolidation method cannot effectively balance the long-term VNF migration cost and energy saving.Therefore,in order to solve the above problems,this paper conducts an in-depth study from the two aspects of VNF deployment optimization and consolidation optimization.The main research contents and contributions are as follows:1)In the process of VNF deployment,reducing the active server can effectively reduce the server energy consumption,but it may cause longer communication paths between VNF,thus increasing the bandwidth resource consumption.How to make a good balance between server energy consumption and bandwidth consumption is a difficult problem when solving VNF deployment optimization.In this paper,an integer linear programming model for online VNF deployment is established which considers minimizing server energy consumption and bandwidth consumption as the optimization goal.Based on this model,a VNF deployment method that optimizes energy consumption and bandwidth consumption is proposed.Considering that the active server with sufficient link resources in the underlying network helps to reduce both energy consumption and bandwidth resource consumption,the proposed method takes in account the link bandwidth resource when deciding switch on a server in the process of each VNF deployment.Experimental results show that compared with existing VNF deployment optimization methods,the proposed method can obtain a better balance of server energy consumption and bandwidth resource consumption.2)In order to improve the time efficiency of VNF deployment optimization method,this paper proposes to introduce deployable domain constraints into the VNF deployment problem,so as to reduce the solution search space.A deployable domain is a small portion of servers in the underlying network.Each set of traffic service requests has a corresponding deployable domain.Deployable domain constraint refers to the fact that when VNF of a traffic service request is deployed,the target server used to deploy VNF can only be searched within the corresponding deployable domain,instead of searching directly among all servers.When constructing a deployable domain,on the one hand,in order to not increase bandwidth resource consumption when searching for the deployment location of VNF in the deployable domain,the server with the shortest path from the flow endpoint is preferred to form the deployable domain of the corresponding requests.On the other hand,in order to avoid affecting the success rate of VNF deployment due to searching for the deployment location of VNF in the deployable domain,the total server capacity in the deployable domain is proportional to the resource requirements of the traffic service requests it serves.Thus,the deployable domain constraints can greatly improve the time efficiency of the VNF deployment optimization method without sacrificing the success rate of VNF deployment and the quality of the solution.3)In data center,applications are usually distributed across multiple VM in different servers.The VM hosting the application business(application VM for short)communicate and collaborate with each other to accomplish their tasks.The application VM deployment optimization problem is how to select the server to host the application VM according to the application VM requirements and available server resources to optimize the required objectives,such as server energy consumption,load balancing,etc.In order to enhance the security and performance of applications,SFC deployment is often required for traffic between application VMs.In NFV-based data centers,the successful deployment of VNF for requests of traffic services between application VMs depends to some extent on where the application VMs are deployed.Therefore,it is necessary to jointly optimize the application VM deployment and the deployment of the required VNF between the application VMs.When the deployable domain constraint is introduced to improve the time efficiency of VNF deployment optimization,the deployment location of VM directly determines the setting of the deployable domain requested by the traffic service between the application VMs,thus greatly affecting the successful deployment of VNF between the application VMs.In this paper,VM and VNF deployment problem are first modeled as a binary integer linear programming model.In order to solve this model,two methods of VM and VNF deployment are proposed.For traffic service requests between VMS,the proposed method takes into account the influence of the deployment location of VMs on the resources capacity of deployable domain,thus greatly improving the success rate of VNF deployment.4)VNF consolidation reduces energy consumption by concentrating VNF into as few servers as possible,but inevitably incurs additional migration costs.Choosing which servers to shut down and then migrate the corresponding VNF is one of the keys to effectively balance energy savings and migration costs.In this paper,a mathematical model of VNF consolidation problem is established,which considers minimizing the long-term average energy consumption and migration cost.For the mathematical model of this problem,a VNF consolidation method based on multi-status characteristics is proposed.This method uses the neural network method to build a policy network,which outputs which servers should be shut down according to the multiple related status characteristics,and then decides which VNF needs to be migrated.In addition,due to the high efficiency of Particle Swarm Optimization(PSO)method in training neural network model,this paper applies PSO method into the training of the policy network.The experimental results show that the proposed VNF consolidation method can save more energy at a lower migration cost.