Research On Service Function Chain Deployment In SDN/NFV Network

With the increase in the number of Internet users and the diversification of services,traditional communication networks have problems such as high cost,difficulty in sharing resources,and insufficient flexibility in network configuration in the face of new service requirements.The emergence of Software Defined Network(SDN)and Network Function Virtualization(NFV)is used to solve the problems in traditional communication networks.Software Defined Network(SDN)is a centralized management and control of network-wide devices through a global controller.NFV decouples network functions from proprietary hardware devices and runs softwarebased virtualized network functions(VNFs)on general-purpose devices,which greatly improves the flexibility and scalability of network function deployment.Therefore,SDN/NFV networks can centrally manage global network devices and resources,improve the flexibility of network service deployment and optimize resource allocation.In the NFV network scenario,network flows pass through some network functions in sequence,which is called a Service Function Chain(SFC).SFC can provide users with specific network services.However,VNFs can be deployed on different underlying physical nodes.In the process of SFC deployment,how to select the optimal node to deploy VNF instances in the underlying physical network nodes,and how to generate deployment paths that satisfy various constraints of SFC,have been It has become a hot spot of current academic research.The resources in the network are often limited.How to efficiently and reasonably allocate network resources to the SFC in the SDN/NFV network to fulfill user requests,reduce network resource costs and improve resource utilization is the current challenge.This paper mainly studies the SFC deployment problem in SDN/NFV network.The main work of the paper is as follows:(1)In the existing SFC deployment work,the SFC deployment is mostly divided into two stages: VNF placement and link placement.This deployment method considers the utilization of network resources is relatively simple.The joint placement of VNF nodes and links also considers the utilization of bandwidth and computing resources,which can effectively improve the utilization of network resources and achieve network load balancing.Aiming at the SFC deployment problem with limited network resource capacity,this paper considers node computing resources and link bandwidth resources as a whole,and takes the minimization of the sum of the cost of the two resources as the optimization goal,and proposes a resource-optimized SFC deployment algorithm Viterbi-RA.By transforming the physical network topology into a logical directed graph,the algorithm selects the path with the least total relative cost that satisfies the constraints according to the link bandwidth resources and node computing resources,and realizes efficient SFC deployment.The simulation results show that,compared with the existing algorithms,Viterbi-RA achieves higher network performance in terms of request acceptance rate,node and link resource utilization.(2)In an SDN/NFV network,SFC deployment needs to consider that providing network services is a dynamic process.SFC requests arrive at the network at different times and leave after staying in the network for a period of time.At the same time,different network services usually have different Qo S constraints,such as bandwidth and latency.Therefore,an adaptive online method is needed to automatically deploy SFC requests with different requirements.To address the above challenges,this paper proposes an adaptive SFC deployment algorithm DRL-Deploy based on deep reinforcement learning.A novel neural network based on graph convolutional networks is employed in the learning agent to automatically extract spatial features in irregular graph topology(i.e.,the underlying physical network).In addition,DRL-Deploy divides the decision into VNF placement decision and link placement decision to solve the problem of large action space encountered when using deep reinforcement learning.DRL-Deploy uses deep reinforcement learning to make VNF placement decisions,and then uses a heuristic algorithm to make link placement decisions.This paper adopts a parallel policy gradient training method to optimize the learning agent while ensuring the validity and robustness of the sample training experience.The results of a large number of simulation experiments show that,compared with the existing algorithms,DRL-Deploy significantly improves the request acceptance rate,long-term average revenue and resource utilization.