Mechanisms For Deployment And Auto-Scaling Of Service Function Chain In Network Function Virtualization

Network Functional Virtualization(NFV)transforms the way we implement network services,from expensive proprietary hardware to inexpensive virtual network capabilities(Virtual Network Function,VNF),It effectively reduces the difficulty and cost of managing and operating the traditional network,and improves the dynamic and flexibility of the network.But there are challenges in enabling efficient resource allocation:first,because VNF arrangements are combined into different service chains(Service Function Chain,SFC)to serve network users.Therefore,the impact and dependency of the VNF changes between the traffic,which makes the deployment of the service chain more difficult.Second,due to the explosive growth of network traffic and mobile devices,as well as the growth of 5G technology,next-generation networks expect to deliver services in an agile,automated,and real-time manner,which places a high demand on the performance of resource orchestration.This paper focuses on the dynamic orchestration of service chain of network functional virtualization.We divide resource orchestration into two phases:the initial deployment phase,the placement of the service chain,and the resource adjustment phase,i.e.the elastic scaling of resources.In the study of service chain deployment,we have innovatively proposed parallel deployment schemes based on deep-enhanced learning.We also use deep-enhanced learning to learn about complex network structures and flow characteristics,choosing the optimal placement location for VNF.Parallelism refers to placing the same VNF in different service chains at the same time,which not only emphasizes the correlation between VNFs,but also improves the utilization of resources.In addition,a heuristic algorithm is designed to allocate resources proportionally to achieve the equalization of server resources and link resource allocation.In the study of the elastic scaling of resources,we propose a hybrid scaling mechanism based on control theory.It consists of two adaptive scale controllers,which can be provisioned on an active basis according to the dynamic changes of load.We make downsizing decisions based on predicted changes in traffic loads to avoid premature release of resources.According to the passive controller's monitoring of server performance and traffic,and the pre-judgment of the active controller,the decision to expand capacity is made.The scaling mechanism can respond quickly to sudden load changes,secure the SLA of the service,consider the heterogeneity of the workload,release resources at the right time,improve resource utilization and avoid oscillations.