Research On Cross-Domain Deployment Algorithm Of Service Function Chain In NFV Environment

There are many drawbacks in the traditional network architecture.With the proposal and development of Network Function Virtualization(NFV)technology,the decoupling of hardware and software of network equipment is realized,which greatly reduces the operating cost,and has received extensive attention from all walks of life.As a product of NFV technology,multiple different types of virtual network functions(VNF)are arranged on demand to form a service function chain(SFC).The emergence of SFC provides network service providers with a more flexible way to meet user service requests.However,due to the constraints of dynamic resource requirements and geographical location,the deployment of SFC is facing great challenges,which is the focus and difficulty of current research.On the one hand,the existing VNF allocation strategies cannot meet the dynamic resource requirements,which can easily lead to the problem of resource over-allocation or under-allocation;on the other hand,most of the current SFC deployment research focuses on single-domain network,and a few deployment algorithms in multi-domain network fail to consider the multi-domain network information isolation problem,the fair and efficient problem of chain-cut allocation,and the multiobjective optimization problem of deployment.Therefore,it is crucial the research on how to provide resource on demand for different types of VNFs,how to ensure that the SFC chain cutting process is more fair and efficient,and how to jointly optimize load balancing and energy consumption when deploying SFC sub-chain in the single-domain network.In this regard,this paper mainly studies these two problems and proposes corresponding solutions.The main technical innovations are as follows:(1)In view of the problem that the previous VNF resource allocation strategy cannot meet the dynamic resource demand in the network slicing scenario,this thesis proposes a VNF resource demand prediction method based on the Two Stage Algorithm(TSA).Firstly,the candidate feature sets that are highly related to the prediction target are selected based on the data features,and then the candidate feature sets are further screened by the greedy forward search strategy to obtain the optimal feature set,and finally different types of prediction models are trained.Simulation results show that the model trained based on this method can achieve better prediction performance,and its prediction error can be as low as 0.051.Compared with the benchmark study,the prediction performance improves by at most about 25%.At the same time,the scalability of the method is good,it is not limited to specific type of VNF and ML algorithm,and the trained model can be directly integrated into the existing SFC deployment algorithm for application,which lays a good theoretical foundation for how to cut the chain of SFC in the next step.(2)In the multi-domain network environment,due to the limitation of geographical location,the deployment of SFC must span multiple network management domains.This thesis proposes an improved SFC distributed cross-domain deployment algorithm.The VNF resource demand prediction model is integrated into the deployment algorithm,and then the SFC is divided based on the remaining resource of the domain network and the demand for different types of VNF resource requirements.Finally,considering the load balancing and energy consumption of the single-domain network,the deployment of the SFC sub-chain in the single-domain network is completed.Simulations are carried out in stages and the results show that in the SFC cutting stage,compared with the average chain cutting and weight chain cutting algorithms,the improved weight chain cutting algorithm proposed in this thesis can achieve better multi-domain network load balancing effect,and the load balance is as low as 19.43%;In the deployment stage of the SFC sub-chain domain,compared with the random deployment and the first fitting deployment algorithm,the SFC sub-chain deployment algorithm proposed in this paper can reduce the energy consumption by up to 9.4% while optimizing the load balancing effect,which is more in line with autual needs.