Research On Methodology Of Virtualized Network Function Orchestration

The diversified service demands from people-to-people,people-to-things and things-to-things communications in future network raise new challenges of 4G network architecture.To decompose network element functions from expensive dedicated hardware,telecom network operators introduce network function virtualization(NFV)in 5G mobile network architecture.NFV uses general purpose hardware and virtualization technology to implement software-based mobile network elements,enabling the operators with the capability of rapid service deployment and flexible network management.Based on the flexible network architecture of NFV,operators propose two novel network service provisioning technologies,network slicing and service function chaining.Such technologies can create different vitualized logical network on homogeneous infrastructure to provide customized network services for the diversified vertical industries.To improve the efficiency of NFV infrastructure resources utilization,this paper relies on National Natural Science Foundation of China and National Science and Technology Major Project to conduct researches on virtualized network function(VNF)orchestration.Resorting to optimization theory,heuristic algorithm and artificial intelligence technology,we carry out our research on two aspects: VNF deployment and VNF scaling,with the objective of improving some key performance metrics,e.g.,resource utilization,service reliability and service scheduling time.The detailed contributions of this paper are as follows.(1)One-stage VNF deployment approach towards minimum scheduling timeExisting researches on VNF deployment problem mainly applied a two-stage optimization model in which the two stages including VNF forwarding graph embedding and VNF scheduling are independent.Such approaches ignored the inherent restriction between VNF embedding and scheduling strategy,narrowing the feasible region in the solution space of VNF deployment problem.To optimize the VNF deployment strategy,we build a one-stage optimization model based on integer linear programming and design a multilayer encoding genetic algorithm to solve the NP hard optimization problem.By encoding the execution order and embedding relationship in chromosomes,we cut down the constraints in evolving process,thus reducing time complexity of the proposed algorithm.Furthermore,we introduce the transmission latency optimization in VNF scheduling problem and design a distributed method of bandwidth allocation via Nash bargaining solution,which further reduces the service scheduling timespan by diminishing the transmission latency between VNF instances.The effectiveness of our heuristic algorithm is verified through numerical evaluation.Compared with existing works,our method shows a better performance of scheduling timespan and reduces the CPU time consumed by algorithm execution in the simulated scenarios.(2)VNF forwarding graph embedding approach based on pooling backup strategy.To reach the tradeoff between service function chain reliability and NFV infrastruture resource utilization,we design a VNF forwarding graph embedding approach based on pooling backup strategy.Compared to the existing joint backup strategy,pooling strategy can adaptively change the resource allocation scheme of backup VNF instances according to resource utilization and reliability requirement,thus improving the flexibility of backup resource management.Based on the designed pooling forwarding graph,we propose a traffic-aware dynamic embedding algorithm via mixed integer programming,which can adjust the number of VNF instances in the backup resource pool according to the fluctuation of real-time traffic.The experiments show that compared to existing joint backup strategy,the proposed pooling backup approach can increase the reliability of service function chain and the ratio of accepted requests as well as decrease the resource cost of VNF embedding.(3)Adaptive virtualized network function scaling based on resource capacity demand predictionIn order to realize elastic scaling of VNF resource capacity,an adaptive VNF scaling method is proposed.First,we design an approach to predict resource capacity demand of requested VNFs based on long-short term memory recurrent neural network and forward neural network.Compared to existing methods,the proposed approach introduces traffic fluctuation,resource demands variation and service type as input features,reducing the negative impact of inaccurate traffic forecasting.Furthermore,based on the proposed approach of VNF resource demand prediction,we use quadratic assignment problem to build VNF resource demand graph embedding model and propose a dynamic encoding genetic algorithm to solve the NP hard problem.The proposed algorithm can adaptively adjust the embedding strategy of VNFs in the resource demand graph,realizing centralized deployment of service function chains.The experiment results show that compared with existing scaling methods,the proposed scaling method can reduce the relative error of resource demand prediction as well as the total number of servers occupied by requested VNF instances.(4)Towards latency-optimal placement and online scaling of monitoring function in MECThe popularity of pervasive computing application promotes the development of Multi-access Edge Computing(MEC).MEC promises to provide sufficient computing capacity close to users and realize smart management at the edge of mobile network.To achieve the aforementioned objectives,it is indispensable to implement real-time monitoring of the whole MEC network.However,the geo-distributed deployment of MEC infrastructure dramatically increases the communication latency of gathering the state information(e.g.current status and resource utilization)from servers at the edge.We address latency-optimal placement and online scaling of monitoring functions in MEC.First,we formally formulate latency-optimal placement of monitoring functions as an integer linear programming problem and proposed a genetic algorithm-based meta-heuristic to obtain the optimal solution with fast convergence.Moreover,to serve the time-varying demand on VNF sresource capacity from diversified mobile services,an online VNF scaling scheme is designed for realizing on-demand resource allocation.The effectiveness of our heuristic algorithm is verified through both numerical simulation and experiments in real cloud environment.Experimental results demonstrate performance superiority of the proposed approach over the state-of-art researches,in terms of algorithm CPU time,total network latency and long-term scaling cost.