A Markov Model of Slice Admission Control

Abstract

The emerging feature of network slicing in future Fifth Generation (5G) networks calls for efficient slice management. Recent studies have been focusing on the mechanism of slice admission control, which functions in a manner of state machine. This paper proposes a general state model for synchronous slice admission control, and proves it to be Markovian under a set of weak constraints. An analytical approximation of the state transition matrix to reduce computational complexity in practical applications is also proposed and evaluated. Index Terms-5G, network slicing, network operations and management, network function virtualization I. INTRODUCTION N ETWORK slicing [1] has been considered as an essential feature and one of the most important enablers of the Fifth Generation (5G) cellular communications networks. It allows mobile network operators (MNOs) to manage and utilize their physical and virtual network resources, i.e. the network infrastructure and the capacity of virtualized network functions (VNFs), in the form of logically independent virtual mobile networks, a.k.a. network slices. It provides broad improvements of scalability, flexibility, accountability, shareability and profitability to cellular networks [2], [3].One emerging challenge brought by network slicing is to efficiently allocate network resources over different slices towards better utility efficiency. More specifically, there are two typical scenarios of such inter-slice resource management. First, when an MNO directly provides services to end users and maintains these services on its own scalable slices, as proposed in [4]. Second, in the resource-sharing use case of socalled Slice as a Service (SlaaS), which is discussed in [5], an MNO packs its resources into standardized atomic slices and rents them to external tenants such as virtual MNOs (VMNOs) and service providers for agreement-based periodical revenue. In both scenarios, the MNO aims to maximize the overall network utility rate (e.g. the revenue rate) by adjusting its resource allocation subject to the constraints of resource pool limit and regulation rules.Compared to the case of MNO's own service optimization, the SlaaS problem is more challenging due to the stochastic nature and fluctuating behavior of tenant demand for resources. Recently, multiple studies have been carried out on this topic [6]-[8], applying the similar framework where a binary decision is made by the MNO for every slice admission, i.e. to accept or decline the tenant request for a new slice. Most of these work consider the system as a state machine, where state transitions are triggered by the MNO's responses to randomly arriving tenant requests.

A Survey on Slice Admission Control Strategies and Optimization Schemes in 5G Network

“…For instance, a virtual gamer may request to be admitted into an eMBB slice, while bandwidth demands may be guaranteed, low latency constraints are difficult to guarantee here, hence, such request may be automatically rejected by an admission control algorithm. On the other hand, in inter-slice admission control, each slice request is evaluated independently and only admitted when the InP can guarantee resource allocation, SLA and QoS [7].…”

“…Random slice request admission may be adopted in order to reduce unfairness during admission control [7], [21]. It achieves a fairly normal distribution over a long period.…”

“…Although machine learning methods may occupy extended duration during training and occasionally require retraining, good results are still achievable. Successive convex approximation (SCA) and alternating direct method of multipliers (ADMM) have also been applied for optimal slice admission control [7], [16]. Optimal algorithms are more complex to design, and ample tuning is required in order to achieve better results.…”

“…A slice admission control algorithm is essential for efficient management of network resources in the 5G network. The slice tenants (STs) or virtual network operators (VNOs) (who are the third-party service providers and own no physical network); control, manages and sell virtual services and the infrastructure network providers (InPs) (who owns and operates the physical network), rely on slice admission control to achieve their predefined objectives [7].…”

Tenant-Aware Slice Admission Control Using Neural Networks-Based Policy Agent

“…The service providers in turn, offer revenue for the resources allocated to their deployed services in their dedicated slice. The network slicing concept has been considered at different scales, abstraction levels and in different network segments in the context of multi-tenant 5G networks [3,6,13,15]. However, regardless of how network slices are defined, they are eventually mapped onto a shared network infrastructure and must be managed by InP to optimize both resource utilization and revenue generated from the deployed slices.…”