The Research On Intelligent Reconfiguration Mechanism Of Network Slice

Recently,with the rapid development of 5G technology,how to use SDN and NFV technology in mobile networks to adopt network slice to meet the differentiated service requirements of diverse services in the network has become a hot research issue.At the same time,with the introduction of network slice,the role of service providers has been separated.On the one hand,due to the limited resources at the bottom,how to reasonably allocate resources to multiple slices directly affects the interests of consumers at the upper slices.On the other hand,after the slice is deployed,the total available network resources of the slice often remain unchanged over a longer time scale.However,within the slice of the network,as the flow arrives and leaves,the slice needs to be dynamically adjusted.The internal resource allocation strategy can meet the needs of dynamically changing services in the network.In a resource-constrained network,in order to maximize the quality of user service and the interests of slice consumers,it is extremely challenging to study the reconstruction of slice resources.The main research contents of this article are as follows:The first part studies the dynamic allocation of bandwidth resources and node processing resources within the core network slice in a short-time-scale scenario with flows arrival / departure.With the goal of maximizing user satisfaction,this problem is modeled as a Markov Decision Process(MDP).According to the theory of reinforcement learning,we use Actor-Critic algorithm to solve this problem,and design the corresponding dynamic optimization allocation algorithm of slice resources.The simulation experiments verify the effectiveness of our proposed algorithm.The numerical results show that compared with the traditional resource allocation algorithm,our proposed AC algorithm can make the flow get better service quality.The second part takes the network slice as a whole,considers the interaction between multiple flows,and proposes an intelligent dynamic reconstruction strategy based on Multi-Agent Reinforcement Learning.As different flows may pass through the same paths or nodes,which involves the preemption of bandwidth resources of the same paths or the preemption of VNF instance resources of the same nodes,it is modeled as a multiuser random game using Multi-Agent Reinforcement Learning theory to solve,and based on this,a distributed multi-agent reinforcement learning algorithm is proposed.Finally,through simulation experiments,the effectiveness of the proposed reconstruction algorithm is verified,and compared with the traditional typical algorithm,our proposed algorithm can maximize the overall satisfaction of the flows within the slice.The third part considers how the slice resources of different operators should be allocated from the perspective of the overall network system,which can maximize the overall interests of slice consumers while protecting the privacy of each operator's data.A federated learning method is used to build a virtual global model OM between different operators.This virtual model does not need to export all the data from each slice.Each slice has autonomy over its own data,allowing the collected data to be saved.Locally,the DQN algorithm is used to train the local model independently,and only the global and local parameters need to be passed and updated under the encryption mechanism until each parameter is basically unchanged.Finally,the method of data decentralized training is verified through simulation,which not only improves the efficiency,makes the performance of the entire algorithm better,but also protects users' privacy.