Research On 5G Core Network Resource Allocation Based On Pointer Network

5G mobile network technology(The Fifth-generation Technology for Mobile Network)changes the traditional network business model,SDN(Software Defined Network)and NFV(Net Functions Virtualization)realize the decoupling of physical infrastructure and network services and make network slicing possible.As one of the key technologies of 5G,network slicing is defined as a network configuration that allows multiple virtualized and independent networks to be created on top of a common physical infrastructure.Each network slice allows organizations to accommodate specific application requirements on the unified network.Based on the infrastructure architecture of SDN and NFV,network slices can be regarded as a service function chain composed of a group of network function blocks that meet user service requests in a certain order and resource allocation problem can be modeled as the embedding problem of the underlying physical network and virtual service network.Therefore,this paper will carry out relevant research on network slicing technology in the context of virtualization technology.For the low latency and high reliability scenario,one of the three major 5G application scenarios,the resource allocation problem is modeled based on the characteristics of the core network sub-slices,and the following optimization works are made.First,this paper proposes a two-step mapping algorithm MPNP for nodes and links to handle this problem.The pointer network is combined with the multi-head attention mechanism and the location embedding layer using the reinforcement learning method are integrated to implement the nodes embedding strategy,then the links embedding strategy is implemented based on the K-shortest path method.This algorithm will find each virtual service request an embedding scheme with the lowest latency to achieve the reasonable allocation of resources.Through simulation experiments,the improved pointer network scheme is compared with the origin pointer network algorithm and heuristic scheme for time-delay placement constraint Equilibrium,and it verifies that MPNP can effectively reduce the average service delay and improve the deployment success rate,while ensuring a good level of resource utilization.Secondly,during the deployment of the overall network slicing,persistent incoming service requests may cause the response network slice queue,and the scheduling order will affect the overall network service performance.Therefore,to optimize the above resource allocation algorithm,this paper proposes a multi-level and multi-objective service function chain scheduling queue model.The two-level scheduling queue is implemented according to the slice mapping failures,and the service function chain in the waiting queue at the current time is scheduled based on the pointer network integrating multiple heads’ attention.Then this paper combines it with the deployment phase of the service function chain to realize a multi-objective joint resource deployment algorithm MMS and use PPO algorithm to train it.MMS algorithm will consider the total latency of network slices and the remaining survival time of slices in the waiting queue and improve the reliability and fairness under the requirement of low latency.Finally,two sets of comparative experiments are completed.One set is based on the linear weighted sum method to test the MMS algorithm with different weight ratios.According to the average total latency,average remaining survival time,average deployment success rate and the average resource utilization rate,the weight coefficient which can perfect meet the needs of the scenario is selected as the final MMS algorithm.In the other set of experiments,taking the average total target value,average resource utilization rate and average waiting latency as the evaluation indicators,this paper compares MMS scheduling deployment joint algorithm with the FCFS scheduling deployment and the scheduling deployment joint algorithm based on feature importance priority.And the experimental results show that in the low latency demand scenario,MMS has lower average waiting latency.It is better at reducing the average total latency and improving the total resource utilization.Finally,the paper summarizes the research work and contents.It discusses and considers the problems encountered during the research work and looks forward to the direction and content of the future research work.