Research On Resource Optimization Allocation Mechanism For Flexible Network Slicing

Based on the fifth-generation mobile communication technology(5G,fifthgeneration),various new application scenarios have emerged and the industry scale has expanded rapidly.In order to support a variety of types of services and meet the needs of the vertical industry,the use of network slicing in complex demand scenarios has become an industry consensus.The Network slicing technology is based on two major technologies: SDN and NFV.In particular,network slicing virtualizes the underlying physical network and allocates network resources appropriately according to the requirements of different business scenarios,thereby creating various network functions.As a consequence,network service providers deploy network slices on the access network and the core network of mobile communication networks according to the needs of users.To be more specific,they provide on-demand network transmission services for different needs of different types of users through access network slices and core network slices.Since network resources are limited,especially for the communication and computing resources in specific coverage areas.How to reasonably allocate available network resources among different network slices,and how to optimize the allocation of slice resources according to the different needs of users within a network slice,will have important impacts on the service performance of the network.In addition,it is necessary to plan the resource allocation of network slices,to maximize the utilization efficiency of network resources and improve the service capacity and quality of the network under the premise of minimizing the reconfiguration cost,because corresponding network management cost and control cost are inevitable in the process of dynamic resource allocation between and within network slices.This paper studies how to improve network service performance through dynamic flexible network slice deployment and management,aiming at the challenges posed by the differentiated service needs of diversified users in mobile communication networks based on network slicing.Specifically,this paper analyzes the interaction and influence of dynamic resource allocation between and within slices from the perspectives of access network slices and core network slices.Based on the idea of hierarchical reinforcement learning,this paper studies the dynamic network resources optimal allocation algorithm between slices and within slices to achieve flexible deployment and intelligent configuration of network slices.The contribution of this paper is as follows:(1)Research and analysis on the mechanism of resource optimal allocation in mobile communication network based on network slice.Firstly,we introduce the related concepts,main technologies of network slicing,and the current research situation on resource optimal allocation mechanism in mobile network based on network slicing from the perspectives of access network and core network.Then this paper analyzes the difficulties and problems in the optimal allocation of network slice resources,which further demonstrates the research significance and direction of this paper.(2)This paper studies the optimal allocation of wireless transmission resources in radio access network slicing.Particularly,in the radio access network based on network slices,the optimal allocation of radio transmission resources could be regarded as a joint optimization problem of resources between and within network slices.Based on the Option-Critic idea,the resource allocation between slices is modeled as a SMDP problem,which can be solved by DQN reinforcement learning framework.The dynamic resource allocation within slices is modeled as an MDP problem and solved by AC algorithm.On this basis,the joint optimal allocation algorithm of radio transmission resources between and within slices in access network can be designed.(3)This paper studies the optimal allocation mechanism of transmission resources for service flow in core network slicing.In the core network based on network slices,the VNF resources and the bandwidth resources also need to be allocated to network slices supporting different types of services firstly.In addition,according to the requirements of different service flows in slices,the optimized mapping and deployment of service function chains(SFC)corresponding to different service flows in slices to physical network nodes and links are completed.Therefore,based on H-DDPG hierarchical learning framework,the dynamic resource allocation between slices is also modeled as a SMDP model.The dynamic deployment of SFC corresponding to the service flow in the slice is modeled as an MDP model.The network environment in core network slice is relatively complex,in the sense that,the layered reinforcement learning model faces a large state and action space.In order to improve the convergence performance of the learning algorithm,DDPG method is used to solve the reinforcement learning model of each layer.Consequently,it's feasible to jointly design the optimal resource allocation scheme of inter-slice and intra-slice.