Research On Key Technologies For Management And Optimization Of End-To-end Network Slicing

The widespread application of 5G in emerging scenarios such as the Internet of Things and connected vehicles has led to a rapid increase in the number and types of network slices.The trends of ubiquitous network coverage,intelligent network slicing,and personalized multi-tenant services pose challenges for endto-end resource management and optimization of network slices in transport networks.In the context of large-scale network slicing,the unified,flexible,intelligent,and efficient allocation of network resources becomes a major challenge in the control and optimization of end-to-end network slicing in transport networks.In this paper,we address this challenge by focusing on resource allocation for end-to-end network slicing,including end-to-end resource allocation for single slices in dynamic heterogeneous networking,resource orchestration for flexible isolation and efficient utilization among multiple slices,and resource co-provisioning for multi-tenant services in dynamic slicing scenarios.The main research contributions and innovations are as follows:(1)We propose a centralized resource allocation scheme for end-to-end slicing in dynamic heterogeneous networks with constraints on inter-domain dynamic links.To address the challenges of centralized control and dynamic constraints brought by the evolution of ubiquitous network coverage in dynamic heterogeneous networking,we design an end-to-end heterogeneous network slicing control architecture.We establish an end-to-end resource allocation model for slices with dynamic elastic services in cross-layer multi-domain dynamic heterogeneous networks,and develop an end-to-end slicing resource allocation method with wavelength fragment awareness.Based on the centralized control of end-to-end network slicing,we analyze the characteristics of dynamic links in heterogeneous networks and the constraints of services on dynamic links.We then solve the resource allocation problem using an improved ant colony algorithm.Simulation results demonstrate that the proposed end-to-end network slicing resource allocation scheme achieves over 24.48%improvement in network capacity compared to non-end-to-end allocation schemes,and exhibits advantages in overall service latency,wavelength fragments,and load balancing.We also build an end-to-end network slicing control platform to validate the control mechanism in heterogeneous networks across layers and domains.(2)We propose an efficient multi-slice resource orchestration algorithm for customized resource isolation requirements under simplified error probability models.To address the contradiction between the uncertainty introduced by network slice traffic prediction errors and the deterministic performance guarantee of the network,we establish a probabilistic feature model for the mixed isolation requirements of multiple slices and traffic prediction in transport networks,and develop a minimum network resource orchestration method with guaranteed resources under personalized isolation requirements.By analyzing and simplifying the features of error probability models,we reduce the feasible solution space and achieve step-by-step and efficient solving of multislice resource orchestration.Specifically,we use dynamic programming to realize an efficient optimal slice grouping algorithm,employ linear feature approximation for fast iterative algorithms within groups,and address resource scarcity and computational complexity issues through the analysis of model priorities and the parameter smoothness characteristics of slice grouping models.Simulation results demonstrate the flexibility and efficiency of the multi-slice resource orchestration method,and experimental verification is conducted on the mixed isolation mechanism of slice resources using the end-to-end network slicing control platform.(3)We propose a multi-tenant resource co-provisioning method for proactive perception of slice utility in dynamic slicing services.To address the contradiction between the uncertainty of multi-tenant active resource coprovisioning in dynamic services and the maximization of slice utility,we establish a multi-tenant active network resource co-provisioning model based on slice value.By analyzing the service value of network slices,we solve the problem of maximizing slice utility based on the active network slicing mechanism.We divide this problem into two steps:active network slicing and slice utility perception.Specifically,we design feature extensions,optimal model selection,and online training and prediction for slice traffic prediction by analyzing slice traffic features,thus implementing the active network slicing mechanism.We also analyze the impact of prediction uncertainty on slice value and design a slice value conversion strategy.The problem is solved using a greedy matching algorithm.Simulation results show that the multi-tenant resource co-provisioning method based on active perception of slice utility effectively improves network resource utilization and maximizes user and tenant benefits.The design and validation of the active network slicing mechanism are also conducted on the experimental platform.