Research On Virtual Resource Management Technology Of 5G Network Slice

Mobile communications will soon enter the 5G era,and 5G will have higher user experience rate,million level connection number density,millisecond end to end delay and other key indicators,as well as a variety of application scenarios.In order to achieve the above vision,we need to innovate the network architecture and business offering mode besides innovating the 5G air port technology.In terms of network architecture,5G introduced the virtualization technology of network function,the flexible deployment of Virtual Network Function(VNF)on universal hardware platform is realized.In terms of business providing mode,5G provides flexible and diversified customized services to users through on-demand network slicing.The virtual network architecture of 5G provides the feasibility for the implementation of the network slicing.In the different application scenarios of 5G,the network slice with different service delivery capabilities can be constructed.The virtual resource management scheme of the network slice will directly affect the efficiency of the 5G service.In the current virtual resource management scheme of network slice,the network slice is usually abstracted as a virtual network or a service function chain,and the research of virtual resource management scheme separately does not consider the unique business attributes and characteristics of 5G,and lack of systematic design,resulting in low resource utilization and increased service delay.Based on this,this paper relies on the national high-tech research and development plan(863)"5G wireless dense network architecture and key technology research and development" and "5G wireless network non stack protocol virtualization key technology research and development",around "how to build a virtual resource management scheme to provide requirements for 5G network slicing services",in accordance with the virtual resource arrangement,deployment and scheduling of three links In the specific problem,starting with the analysis of the characteristic attribute of 5G,the key technology is researched,and a set of systematic network slice virtual resource management solutions is put forward,and the following results are obtained.1.In order to solve the problem of multi-dimensional and high order clustering in the case of unknown cluster number,a clustering method of VNF instance for performance compatibility is proposed for the VNF instance clustering requirement of the choreography stage.First,the performance compatibility between VNF instances is defined,and the clustering of multidimensional and high level VNF instances is transformed into hypergraph clustering models.Then,hypergraph clustering model is described by evolutionary game theory,and the clustering process is transformed to search process of evolutionary stable strategy cluster.Finally,a repetitive dynamic equation is designed to achieve the solution of evolutionary stable strategy clusters.The experimental simulation results show that this method can cluster the high performance VNF instances into a cluster.The performance compatibility of VNF in the cluster is higher than 95%,and the clustering error is reduced by more than 18% in the acceptable range of clustering time.The combination time of the VNF increases linearly with the increase of the real number of VNF examples,and is less than the combination time of the traditional algorithm.2.Facing the deployment requirement of cross domain VNF deployment,in view of the problem that the existing cross domain deployment algorithm does not consider the problem of node computing resources and link bandwidth resources,a static deployment method of 5G cross domain virtual network based on DPSO-K is proposed.This method transforms the cross domain deployment of VNF with static resource requirements in 5G network into a cross domain virtual network mapping,and designs a mapping algorithm for cross domain virtual network with the goal of minimizing the resource and link bandwidth resource of the node.Firstly,with the goal of minimizing the bandwidth overhead of inter domain links,the optimization discrete particle swarm optimization(DPSO)is used to complete the selection of inter domain links and boundary nodes,and the optimization ability is improved.Then,the node bidding mechanism is introduced to the mapping in the domain with relatively small overhead,and the minimum resource and link bandwidth resources are minimized by the node,and Kr is used.The uskal minimum spanning tree fast algorithm realizes the VNF deployment in the domain and improves the speed of the business on-line.The experimental simulation results show that the average request acceptance rate of the proposed method is 97%,although the operation time is slightly increased,but compared with the traditional method,the average cost is reduced by more than 6.6%,and the cost is less affected by the number of data domains.3.In order to deploy VNF deployment requirement in the deployment phase,a dynamic deployment method of virtual network function in 5G domain based on Q-learning is proposed in view of the problem that the existing domain algorithms have not implemented joint deployment of nodes and links,and the business and deployment have not been implemented synchronously.This method transforms the domain internal deployment problem of VNF with dynamic resource requirement into service function chain deployment in 5G network,and designs the shortest path deployment algorithm of service function chain with the goal of node processing delay and minimum transmission delay.In this algorithm,the deployment of service function chain is modeled as a Markov decision process,and the Q-learning algorithm is designed to solve the problem,and the VNF deployment scheme with the smallest network service delay cost is dynamically searched.The experimental simulation results show that the average request acceptance rate of the proposed method is above 98%,and the average time delay can be reduced by 2% compared with the traditional method,and the convergence time of the algorithm is the shortest.4.For the low delay scheduling requirement of VNF under the scheduling phase concurrency service request,in view of the existing scheduling algorithms separating the virtual machine selection and scheduling,and without considering the randomness of the VNF parameters,a virtual network function scheduling method based on the complex behavior strategy in the 5G domain is proposed.This method first relaxes the location constraints of VNF during the deployment process,then introduces a complex behavior strategy to describe the scheduling problem in the Markoff decision process,and realizes the combination of virtual machine selection and VNF scheduling.Finally,an enhanced learning algorithm is designed to solve the problem,and the global optimization of resource constraints and service delay constraints is realized.The simulation results show that after the scheduling of this method,the total service delay of multiple SFC is reduced by more than 5.3%,and the shortcoming is that the complexity is relatively high.