Research On Optimal Deployment Of Service Function Chains In NFV-Enabled Networks

In order to solve the problems such as long development cycles,complex management and difficult technological innovation in existing telecommunication networks,the European Telecommunication Standards Institute proposed the concept of Network Functions Virtualization(NFV),which aims to decouple network function software from proprietary hardware,and encapsulate a series of Virtual Network Functions(VNF)into Service Function Chains(SFCs)to provide flexible,scalable and diversified services for users.A series of researches and applications on the optimal deployment of SFC have been carried out by academia and industry at home and abroad,but it still faces two timely burning challenges: high resource consumption and insufficient Quality of Service(Qo S).Therefore,this paper studies the optimal deployment of SFC in NFV-enabled networks,focusing on solving the problems of "collaborative optimization of resources and Qo S for multi-SFC static deployment" and "collaborative optimization of resources and reliability for SFC dynamic deployment".Aiming at the collaborative optimization of resources and Qo S for multi-SFC static deployment,it is critical to effectively deploy multi-SFC into networks with as little resource consumption and as good Qo S as possible.Therefore,this paper proposes a Markov-chain-based SFC static optimal deployment(MSCD)method,which formulates the SFC deployment as a Markov-chain-based model,containing the VNF initial state,the VNF state transition matrix and the link state probability matrix.On this basis,this paper proposes a heuristic method,based on improved Viterbi and Backward algorithms,to reduce the resource consumption during VNF deployment,and improves Qo S when deploying virtual links.It has been proved by simulation results that MSCD can map VNFs and virtual links to the underlying network with high successful deployment ratio for SFC and great load balancing,and also achieves a balance between resource consumption optimization and Qo S guarantee.Aiming at the collaborative optimization of resources and reliability for SFC dynamic deployment,it is critical to utilize dynamic and limited resources in the network to provide reliable services for SFC requests.Therefore,this paper proposes a reliability-aware VNF placement(RVFP)method,via Multi-Agent Deep Reinforcement Learning(MADRL),which adopts VNF backup and improved Multi-Agent Deep Deterministic Policy Gradient(MADDPG)to assign a single SFC into multi agents for cooperative deployment.Each agent aims at maximizing reliability or minimizing failure probability.They conduct intelligent,reliable and dynamic VNF deployment with centralized training and distributed execution through intra-domain cooperation.It has been proved by simulation results that RVFP can effectively train a SFC dynamic deployment method and achieve higher reliability with less resource consumption.