Online Deployment Technology For Service Function Chain Based On Software Defined Network And Virtual Network Function

With the increasing popularity of Software Defined Network(SDN)and Network Function Virtualization(NFV),the integration of SDN and NFV(SDN/NFV)also develops quickly in both academic and industry areas,which makes many contributions in development of network techniques such as software-defined datacenter and 5G network,but also has some performance bottlenecks.In SDN/NFV platforms such as datacenters,a network service is provided as a sequence of Virtual Network Functions(VNFs)executed in a pre-defined order,which is known as a Service Function Chain(SFC).Since datacenter traffic typically exhibits high variations due to stochastic arrival and departure of requests,online SFC deployment is urgently needed so to dynamically minimize the operation cost of occupied servers and maximize the traffic volume of accepted requests.In this paper,we formulate the SFC deployment problem as a Markov Decision Process(MDP)based optimization problem.To automatically and continually adapt to network variations,we devise an efficient Deep Reinforcement Learning(DRL)approach with serialization backtracking and feature engineering(DRLSF)for online SFC deployment.In particular,we introduce a serialization backtracking method to efficiently deal with large discrete action spaces.To learn the dependency relation in the SFC deployment process,we investigate a feature engineering method to add the feature of the subsequent unplaced VNF to the current network state.Extensive simulation results demonstrate that our approach not only provides fast convergence to efficiency and rapid response to dynamic requests,but also surpasses the state-of-the-art solutions by 20.44% higher accepted traffic volume and 16.83% higher profits of internet service providers in average.