Research On 5G Network End-to-End Slicing Algorithms

With the advent of the 5G era,a variety of business scenarios will appear,and different business scenarios need to meet different communication requirements,such as unmanned driving,virtual reality and smart factories.They have all the delays,bandwidth and reliability of communication.There are different requirements.Traditional mobile communication networks often only provide services for the mobile broadband service,which is not suitable for 5G diversified business scenarios.If providing a unique network for each specific business,it will consume huge construction costs and network operation and maintenance will become extremely complicated.To solve this problem,5G proposes a network slicing technology,which divides the traditional physical network into multiple independent virtual networks to provide specific services for different businesses.The emergence of network slicing technology will inevitably bring a major problem to the allocation of wireless resources.For end-to-end network slicing,the access side needs to allocate virtual wireless resources for it,and the core side needs to allocate service links for it.Only by comprehensively considering the two aspects,the end-to-end network slicing wireless virtual resources can be allocated properly and the network performance can be optimized.This paper mainly does the following three aspects of work:(1)Two types of rate-constrained and delay-constrained wireless resource allocation models for slice access side are proposed.Users of different slices have different Qo S requirements,and virtual wireless resources are allocated to each user on the premise of satisfying user Qo S,so that the resource utilization rate of all slice users on the access side is maximized.In the solution phase,the base station is regarded as a backpack with a certain capacity,the user is regarded as an item,the weight of the item is the resource required by the user,and the income is the system capacity.However,it can’t be solved by the conventional knapsack algorithm,because the number of RBs required for users to access different base stations varies.Therefore,this paper adopts a dynamic programming algorithm based on the improved knapsack algorithm to solve the problem.The simulation results show that the algorithm can improve the user’s access rate under the premise of meeting the needs of various slice users.(2)For the network topology of different types of slices on the core side and the business requirements of the corresponding types of slices,each service chain is mapped to maximize user capacity.A model of link mapping on the core side is established.Prioritize the service chains of the same type,perform point-by-point mapping of the service chains in the order of priority,evaluate each candidate node,and select the node with the largest evaluation score to map sequentially until the service chain mapping is completed.The algorithm can map a reasonable link and virtual machine selected by the user and meet the user’s delay and bandwidth requirements.Make the access rate of core users as large as possible.(3)An end-to-end network slicing algorithm based on DQN is presented in the paper,so that the performance of the entire end-to-end network slice is optimal.Use deep reinforcement learning algorithms to adjust slice resources.Use feedback from each environment to dynamically adjust resources to obtain better resource allocation strategies.The results show that considering the static environment and dynamic environment,the algorithm can be gradually trained to obtain a better solution,and the allocation of resources can be adjusted at any time according to changes in the environment.