Resource Allocation Technologies For Network Slicing

With the continuous development of information technology,the connection between the Internet and human production and life has become deeper and deeper,and some emerging network services closely integrated with vertical industries have also continued to emerge.This requires Internet technology to be flexible according to the service quality requirements of various network services to provide customized and high-quality network services.In the context of this demand,network slicing technology that can provide users with end-to-end customized network services has emerged as the times require.This paper will focus on the resource allocation technology of network slicing.The resource allocation technology of network slicing is facing challenges brought about by demands such as flexible allocation,performance isolation between slices,and end-to-end control.Specifically,the elastic allocation of slice resources is often constrained by performance isolation requirements.How to effectively control the risk of damage to slice performance isolation to achieve efficient and elastic resource allocation is one of the urgent challenges to be solved.In addition,the end-to-end feature puts forward higher requirements for the scalability of slice control.How to flexibly schedule the computing resources of the control layer in a wide area to optimize the transaction processing delay is the second challenge to be solved.This paper will study the resource allocation technology of network slicing from three aspects:the dynamic resource allocation problem and slice mapping problem of the challenge one,and the control layer delay optimization of the multi-domain transmission network slicing challenge two.The main contributions are summarized as follows:1.A load-drievn dynamical network slicing resource allocation mechanism(DRA-IG)is proposed.Facing the trade-off between user isolation requirements and resource utilization efficiency in slicing resource allocation,the paper proposes a load-driven dynamic allocation of slicing resources DRA-IG.DRA-IG models the isolation requirement of slices as chance constraints,and takes resource utilization efficiency as the optimization goal,and obtains a slice resource allocation optimization problem based on chance-constrained programming.The method of data-driven distributionally robust optimization transforms the original chance-constrained programming problem with uncertainty into an easy-to-solve deterministic positive semi-definite programming problem based on the second-order moment information of the load.DRA-IG further proposes a secondorder moment information estimation scheme that combines prediction methods and deep reinforcement learning.Simulation experiments show that DRA-IG can effectively regulate isolation requirements and resource utilization efficiency by setting the isolation guarantee level parameters of the slices,and DRA-IG's second-order moment estimation module can effectively correct the estimation error according to the actual state of the slice.2.A Network Slicing Embedding solution(NSE-RRS)supported Risk-controlled Resource Sharing is proposed.In order to effectively improve the resource utilization efficiency of slice mapping and reduce the actual cost of physical resources,the paper proposes a network slicing embedding solution(NSE-RRS)supported riskcontrolled resource sharing.NSE-RRS divides slices into elastic and inelastic according to the type of traffic,and assumes that elastic slices allow a controllable degree of resource sharing.NSE-RRS uses chance constraints to control the possible risks of resource sharing within the setting range,and models the mapping problem that supports sharing as a mixed integer programming problem with chance constraints.Simulation experiments show that NSE-RRS effectively improves resource utilization efficiency and the revenue/cost ratio of slicing requests,and NSE-RRS can effectively adjust the conservative degree of resource sharing through risk control parameters.3.A SDN multi-controller model HHS for slicing control delay optimization is designed.Facing the requirements of end-to-end slicing for the scalability of the control layer,the paper uses transaction processing delay as the main performance index,studies the control layer modeling of multi-domain transmission networks,and proposes a new type of multi-controller hybrid hierarchical structure,named HHS.In the HHS model,the processing controller of any transaction is no longer in a fixed position like the traditional flat and hierarchical model,but can be dynamically configured based on the needs of time delay optimization.This chapter further takes the slice end-to-end channel establishment transaction as a research case.The simulation experiment shows the effectiveness of the HHS model-based processing controller location optimization problem.The controller location can be dynamically set to achieve the optimal transaction processing delay under different request arrival rates.