RESOURCE PROVISIONING FOR NETWORK SLICING IN 5G AND BEYOND

24/3/2025
RESOURCE PROVISIONING FOR NETWORK SLICING IN 5G AND BEYOND

Abstract: Network slicing is a fundamental component of 5G networks, enabling Mobile Network Operators (MNOs) to create tailored slices for Service Providers (SPs) to deliver customized services. These network slices share a common infrastructure owned by an Infrastructure Provider (InP), making efficient resource allocation across slices essential. This paper summarizes the work presented in the thesis (Q.-T. Luu, 2021). Taking the perspective of the InP, this thesis introduces several resource provisioning methods for network slices. Unlike previous best-effort approaches, which deploy the Service Function Chains (SFCs) of a slice sequentially across the network infrastructure, this thesis focuses on provisioning aggregate resources to meet the demands of the slices. Once resources are successfully provisioned, the SFCs within a slice are guaranteed sufficient resources to operate effectively, ensuring that quality of service requirements are met. Additionally, the proposed solutions reduce the computational overhead required for deploying the SFCs.

Keywords: Network slicing, resource reservation, resource provisioning, resource allocation, slice admission control, linear programming, 5G and beyond.

I. INTRODUCTION

A. Context

Beyond providing connectivity, the fifth generation (5G) mobile network presents operators with unique opportunities to embrace new business models tailored for consumers, enterprises, vertical industries, and third-party partners. By targeting various industry sectors, 5G networks facilitate enhanced automation and monitoring. Dedicated services for vertical markets, such as energy, e-health, smart cities, and connected vehicles, can be deployed more efficiently (Li et al., 2017). The 5G architecture is designed to offer the flexibility needed to support diverse services, each with stringent requirements for latency, throughput, and availability (Kaloxylos, 2018).

To achieve this flexibility and enhance adaptability, mobile networks are evolving into systems of virtual resources that can be instantiated and decommissioned on demand to meet customer needs promptly. Key technologies such as Software-Defined Networking (SDN) and Network Function Virtualization (NFV) play an increasingly critical role in enabling this y (Basta et al., 2014).

With the integration of SDN and NFV, network slicing has emerged as a transformative enabling technology technology (5G Americas, 2016; Barakabitze et al., 2020; Lopez & Telefonica, 2017). By enhancing the way networks are managed, network slicing reduces overall equipment and management costs (Liang & Yu, 2014) while increasing operational flexibility (Rost et al., 2017). It allows for the parallel management of multiple dedicated end-to-end virtual networks or slices over a shared infrastructure. This enables vertical industries to be more effectively addressed, as customers can operate their own applications on customized network slices tailored to their specific needs (GSMA, 2018).  

As highlighted in (Weldon, 2015), the networking industry is undergoing a significant transformation toward network virtualization and cloud technology, evidenced by a surge in patent filings, demonstrations, proofs-of-concept, field trials, and commercial deployments. Technologies such as network slicing are instrumental in creating additional value for enterprises, further driving innovation in the industry.

B. Slice Resource Provisioning

Contrary to traditional best-effort approaches, where Service Function Chains (SFCs) of a slice are deployed sequentially within the infrastructure network, this thesis introduces a new approach that provisions resources in advance to accommodate slice resource demands. Specifically, infrastructure resources are reserved beforehand for the future deployment of SFCs belonging to each slice. 

Once resources are provisioned for a given slice, the SFCs are guaranteed access to those reserved resources, ensuring the fulfillment of contracted service requirements with the desired quality. Furthermore, as demonstrated in the subsequent chapters, this resource provisioning method significantly reduces the computational overhead required for SFC deployment.

Figure 1 illustrates the contrast between traditional SFC embedding approaches (Figure 1a) and the proposed slice resource provisioning approach (Figure 1b). In the latter, the SFC deployment process is divided into two distinct phases: first, resource provisioning is conducted for a given slice; second, the SFCs of that slice are deployed within the pre-allocated resources resulting from the first phase.

A diagram of a diagram of a diagram Description automatically generated

Figure 1: Illustration of (a) a direct SFC embedding, and (b) the proposed two-stage approach, where slice resource provisioning is performed before the SFCs deployment within the provisioned resources.

From the perspective of an Infrastructure network Provider (InP), this thesis explores various provisioning frameworks tailored to different use cases. Initially, a slice resource provisioning framework is proposed to address multiple slice demands, considering computing, memory, and wireless capacity. The framework is then extended to scenarios where slices must be deployed over specific geographical areas, incorporating additional constraints such as coverage and minimum per-user rate requirements. Finally, slice resource provisioning and admission control are combined to cope with 

  1. the uncertainties related to the slice resource demands, including factors such as fluctuations in user resource requirements and changes in the number of users over time; and

  2. the dynamic nature of slice requests, characterized by the arrival and departure of slices requests.

The proposed approach aligns fully with the 3GPP framework for managing network slicing (3GPP, 2020). The slice resource provisioning methods introduced in this thesis are designed to operate during the network environment preparation task within the preparation phase, as illustrated in Figure 2. During this phase, activities such as network slice design and capacity planning, on-boarding and evaluation of necessary network functions, and provisioning of infrastructure resources are completed prior to the creation and activation of network slice instances. A detailed discussion of the management aspects of network slicing will be provided in Chapter 2.

A diagram of a diagram AI-generated content may be incorrect.

Figure 2: 3GPP view on network slicing managements aspects [3GPP, 2020].

The rest of this paper is organized as follows. Sec. II presents the research challenges related to the problem of resource provisioning for networ slicing in 5G and beyond. Sec. III summarizes some related work. The main contribution of the thesis is introduced in Sec. V. Finally, some conclusions and perspectives are drawn in Sec. VI.

II. RESEARCH CHALLENGES

The first key challenge is determining how to provision each network slice with an appropriate amount of physical resources (such as computing, storage, and network resources) to meet its resource demands while satisfying predefined service requirements. The amount of resources allocated to a slice is influenced by the services it supports and their Quality of Service (QoS) requirements, which are typically expressed in terms of latency, bandwidth, computing, and storage needs. These requirements, in turn, depend on the demand for the services within the slice. Moreover, it is anticipated that only a limited number of network slice types will coexist, driven by business viability, with examples including ultra-HD video, e-health, sensor networks, intelligent transportation systems, gaming, and tactile internet applications. Better understanding and defining the characteristics of these slices will enable more efficient resource provisioning. With this context in mind, Challenge 1 is formulated.

Challenge 1. Enough infrastructure resources should be provisioned to accommodate slice resource demands, so as the desired service requirements are satisfied. The amount of resources provisioned to a slice depends on the characteristics of the service it provides, its QoS requirements expressed, e.g., in terms of bandwidth, computing, and storage requirements.

Many research challenges remain when network slicing incorporates the wireless part of legacy or 5G networks (Kaloxylos, 2018; Li et al., 2017), where radio access has to be considered. For instance, in (Chatterjee et al., 2018), the service characteristics required by an SP are: the minimum data rate, minimum rate coverage probability, the density of user equipments (UEs), and the geographical zone to be covered by the slice. In this thesis, we also tackle the problem of slice resource provisioning with some coverage constraints. Challenge 2 summarizes such issues.

Challenge 2. In a wireless slicing context, e.g., RAN slicing, some constraints related to the coverage area of the slice as well as the user location also have to be taken into account.

In the survey (Barakabitze et al., 2020) on 5G network slicing, the authors provide a taxonomy of network slicing, architectures, and future challenges. One of the open questions is how to meet the slice requirements of different verticals, where multiple network segments including the radio access, transport, and core networks have to be considered. Infrastructure networks on which slices are operated must support high-quality services with increasing resource consumption (video streaming, telepresence, augmented reality, remote vehicle operation, gaming, etc.). Moreover, the number of users of each slice, their location (usually difficult to predict (Richart et al., 2016)), and resource demands may fluctuate with time. These uncertainties may impact significantly the resources consumed by each slice and make the slice resource provisioning problem more challenging. Enough infrastructure resources should be dedicated to a given slice to ensure an appropriate QoS despite the uncertainties in the number of slice users and their demands. Over-provisioning should also be avoided, limiting the infrastructure leasing costs and leaving resources to concurrent slices. This leads to Challenge 3.

Challenge 3. An efficient slice resource provisioning mechanism should be robust against the uncertainties related to slice resource demands. Moreover, the proposed approach has to be implemented so as to limit its impact on low-priority background services, which may co-exist with slices in the infrastructure network.

In addition to the uncertainty issue, it is also necessary to account for the dynamic nature of slice provisioning requests: Slice requests arrive at different time instants, with various activation delays, life durations, and time-variant resource demands. These parameters significantly impact the aggregate resource demands of network slices. The variety of services supported by slices induces very different QoS requirements (Li et al., 2018). In traditional slice resource allocation approaches (Barakabitze et al., 2020; Su et al., 2019; G. Wang et al., 2017), resources are allocated to slices just before its required activation time. With such a just-in-time slice management, it is difficult to guarantee the availability of enough infrastructure resources at the deployment time and during the life-time of a slice. In such a case, slice demands may be rejected. Therefore, a novel slice resource provisioning approach should be introduced, providing anticipated slice admission control. Slices are admitted, possibly largely before their activation time when enough infrastructure resources are available to meet their QoS requirements. This leads to Challenge 4.

Challenge 4. Slice provisioning requests should be processed in an anticipated way, largely before their activation time, to guarantee the availability of infrastructure resources at the deployment time and during the life-time of the slices. The resulting slice admission control mechanism should take into account the dynamic nature of slice provisioning requests and the priority level of slice requests.

III. RELATED WORK

A. SFC Resource Allocation

A large and growing body of literature has investigated the SFC resource allocation problem. Since a slice can be seen as a collection of SFCs, allocating resources for a given slice means allocating resources for all SFCs constituting that slice.

The SFC resource allocation problem is usually represented as a mapping of elements (virtual nodes or VNFs and virtual links) of SFCs onto the physical infrastructure. The mapped infrastructure nodes and links must satisfy some specific requirements of the virtual nodes and virtual links of the SFCs, e.g., in terms of resource demands, latency, or availability. In the literature, the problem of SFC resource allocation is also called Virtual Network Embedding, VNF placement, or SFC embedding.

In (Riggio et al., 2016; Vizarreta et al., 2017a), computing, memory, and aggregate wireless resource demands of SFCs are considered. The minimization of the SFC embedding cost is formulated either as an Integer Linear Programming (ILP) (Cohen et al., 2015; Riera et al., 2016; Vizarreta et al., 2017b) or as a Mixed Integer Linear Programming (MILP) problem (Chowdhury et al., 2012), which are known to be NP-hard (Fischer et al., 2013). 

To address the high computational complexity resulting from the ILPs or MILPs, various heuristics have been proposed, see, e.g.,  (Cohen et al., 2015; Riggio et al., 2016; Vizarreta et al., 2017c). For example, (Riggio et al., 2016) introduced a heuristic based on the search of shortest paths to sequentially embed the SFCs. In (Vizarreta et al., 2017c), the candidate infrastructure nodes are sorted to find the best node, in terms of deployment cost, to host a given VNF. Its neighbors are then considered as candidates to deploy the next VNF.

The Column Generation (CG) technique has been widely studied to solve large ILP problems (Huin et al., 2017a). With CG, the original ILP is decomposed into a Master Problem (MP) and a Pricing Problem (PP). The MP is the original problem where only a subset of variables is considered. The PP is a new problem created to identify a new variable, i.e., a column, to add to the MP to improve the current solution. In (Huin et al., 2017b) or (Liu et al., 2017), CG has been used to relax ILP-based SFC embedding or reconfiguration problems. Specifically, in (Huin et al., 2017b), the SFC embedding problem is addressed. Only core capacity and bandwidth resources for infrastructure nodes and links are considered.

The resource allocation among multiple slices is performed considering two different approaches. The first involves a centralized convex optimization problem, whose objective is to maximize the total slice utility. Nevertheless, as pointed out in (Halabian, 2019), such centralized solution lacks scalability, is not robust to a failure of the central optimizer, and is prone to non-collaborative slice providers who may harm the system. For these reasons, a distributed method based on game theory is considered to improve robustness and scalability. Optimization is performed in a decentralized way among the data centers and slice providers. The results provided by all entities determine the final resource allocation for all slices. Nevertheless, the placement of VNFs in data centers is predetermined by the MNO and, again, wireless resources are not considered.

B. Coverage-Aware Slice Resource Allocation 

The design of efficient allocation mechanisms for virtualized radio resources has been recently addressed in (Chatterjee et al., 2020) so as to meet SP demands, while providing, with a given probability, a minimum data rate for any user located in their coverage area. The rate constraint is expressed as a linear function of the BS load (number of users served by the BS), of the distance from users to the nearest BS, and of the downlink interference. This linear approximation, however, requires some assumptions. For instance, a user of an SP is assumed to be served by its nearest BS among the set of BSs allocated to the SP. This reduces somehow the potentiality of achieving the optimal sharing of the radio resource.

In (Teague et al., 2019), a heterogeneous spatial user density is considered, and the joint BS selection and adaptive slicing are formulated as a two-stage stochastic optimization problem. The first stage aims to define the set of BSs to activate. The second stage aims at allocating wireless resources of the BSs to each point of the region to be covered by the SP. Several random realizations of user locations are generated to get a reduced-complexity deterministic optimization problem. A genetic algorithm is then used for the optimization.

In (Lee et al., 2016), a network slicing framework for multi-tenant heterogeneous cloud radio access network is introduced. The sharing of radio resources in terms of data rate is considered, with some constraints related to the fronthaul capacity, the transmission power budget of RRHs, or the tolerable interference threshold of an RRH on a sub-channel. Slicing is formulated as a weighted throughput maximization problem, which aims at maximizing the total rate obtained by users connected to given RRHs on given sub-channels. Nevertheless, the proposed framework does not consider computing and memory resources associated to the processing within the BBUs. Such resources are assumed to be properly scaled so as to support the required service rate. Moreover, the proposed framework addresses only downlink data services.

A game theory-based distributed algorithm is proposed to solve the problem of wireless network slicing in (D’Oro et al., 2019). The proposed algorithm accounts for the limited availability of wireless resources and considers different aspects such as congestion, deployment costs, and the RRH-user distance. The coverage area of RRH is considered, but the possible coverage constraints required by different slices are not considered.

C. Uncertainty-Aware Slice Resource Allocation

In many conventional approaches, enough network resources are allocated to make a service available to all users, all the time. Many applications such as e-mail and instant messaging do not require such exclusive service. To address this problem, in (Trinh et al., 2011), flexible service availability levels are defined. These flexible levels lead to cost savings for the infrastructure provider that can offer overbooked resources for users accepting a service with possibly degraded availability. In the context of network slicing, SPs can benefit from such an approach by providing services with reduced availability or degraded quality to some users ready to accept these conditions. Nevertheless, to evaluate the incidence on the QoS of such under-provisioning mechanism, it is necessary to introduce models of the number of users of a service and of the resource consumption. Such models have not been considered in (Trinh et al., 2011).

A worst-case allocation at peak traffic is considered in (Huin et al., 2017b; Y. Wang et al., 2017). Nevertheless, this infrastructure resource overbooking is costly and most of the time unnecessary, as all individual slice resource demands are very unlikely peaking simultaneously. In (Coniglio et al., 2015), the virtual network embedding problem is solved considering uncertain traffic demands. An MILP formulation is considered, where some of the constraints are required to be satisfied with high probability. In (Mireslami et al., 2019), the total deployment costs for cloud computing applications are minimized, while satisfying some QoS constraints. To cope with the uncertain nature of the demands, a stochastic optimization approach is adopted by modeling user demands as random variables obeying normal distributions. Deployment is performed based on the mean demands increased by an integer amount of their standard deviations. This might lead to a conservative solution, requiring more allocated resources than needed. This also reduces somehow the possibility of having service-dependent satisfaction levels.

A network slice embedding problem is considered in (Fendt et al., 2019), where available resources and resource demands are assumed to be partly uncertain. They are described by normal distributions built upon the data history on mobile network resource availability as well as slice resource utilization. To control the probability that a slice embedding solution will benefit from enough infrastructure resources, despite the uncertainties, some adjustable safety factor is introduced. As in [(Mireslami et al., 2019), enough resources are dedicated to a service so as to satisfy the mean plus times the standard deviation of the demands. In (Fendt et al., 2019), additionally, a similar approach is considered to account for the uncertainty in the available resources. A probability of feasibility, depending on , is then evaluated for the slice embedding to measure the risk of having a degraded service for some users. The proposed solution leads to a slice resource allocation solution robust to uncertainties. Nevertheless, the resource demands of the different components of the slice are often considered as independent. Moreover, the safety factor is usually the same for resource demands and available resources. This overestimation may require more resources than strictly necessary and increases the operation cost.

The network slice embedding problem with demand uncertainties is also addressed in (Baumgartner et al., 2018a). The minimization of deployment costs considering first static resource demands is formulated as an MILP. Two robust network slice design formulations are then proposed to (i) handle demand uncertainties, and (ii) additionally account for correlations among the uncertain demands. A tuning parameter is introduced to control the trade-off between robustness to the demand uncertainties and the deployment costs. Uncertainties related to the background traffics on the infrastructure, which clearly affect the residual infrastructure resources, are not considered.

To reduce the computation effort required to solve the robust network slice embedding problem, (Bauschert & Reddy, 2019) proposes to use a genetic algorithm, shown to surpass the performance of state-of-the-art robust MILP solvers used, e.g., in (Baumgartner et al., 2018a). Uncertainties in infrastructure link bandwidth are also considered in (Wen et al., 2019), where possible failures of infrastructure nodes or links are taken into account to propose a robust algorithm that minimizes the network resource consumption under uncertain demands, while remapping the network slice in case of infrastructure failures. Since (Baumgartner et al., 2018b), (Bauschert & Reddy, 2019), and (Wen et al., 2019) assume that the distribution of the variable demands and available infrastructure resource are unknown, their optimization are relatively conservative. Furthermore, uncertainties in various types of resources such as computing, memory, or wireless are not addressed.

D. Admission Control with Dynamic Slice Requests

The topic of dynamic slice/SFC deployment has also received significant attention in recent literature. For instance, in (Liu et al., 2017), a dynamic resource allocation for SFCs is investigated. The deployment of newly arrived SFCs and readjustment of in-service SFCs are taken into account. An ILP formulation is used to address the dynamic deployment problem, aiming at minimizing the cost of VNF deployment and migration. A pre-calculation of all possible routing paths has to be performed in advance, which requires some computational effort before using the deployment algorithm. In (G. Sun et al., 2019), the adaptive adjustment of allocated resources of each slice is enabled after each decision time period (slicing time). An hybrid slice reconfiguration framework is introduced in (G. Wang et al., 2019). The slice can be reconfigured either within small time intervals for individual slices, or within large time intervals to readjust resource allocation of multiple slices. A deep-learning approach is adopted in (Huynh et al., 2019) for dynamic slice resource allocation, with the aim to maximize the long-term revenue of the network provider. Uncertainties related to the slice allocation requests and occupation time are considered. Nevertheless, each slice is assumed to be indivisible, i.e., not made up of multiple elements (e.g., VNFs), which somewhat over-simplifies the problem of slice resource allocation.

Slice admission control (SAC) mechanisms have been developed recently (Bega et al., 2017, 2020; Ebrahimi et al., 2020; Han et al., 2020; Noroozi et al., 2019) to address issues related to the unavailability of enough resources to satisfy all slice requests. In (Noroozi et al., 2019), SAC is formulated as a boolean linear program and a two-step sub-optimal algorithm based on variants of the knapsack problem are proposed to alleviate the complexity. Admission is done for slices with the highest profit considering first the RAN and aggregate core network resources. In a second step, the core network resources are considered without any aggregation to determine whether a slice deployment is possible.

In (Ebrahimi et al., 2020), SAC and resource allocation are performed jointly, to minimize the power consumption of the cloud nodes and the network bandwidth of the infrastructure provider. Transmission delay is taken into account in the slice SLA. Some elastic variables are introduced in an ILP formulation to extend the bounds on some constraints. They help determining when resources may be lacking, in which case slices are rejected starting from those with the highest requirements in terms of resource. Nevertheless, the dynamics of slice requests (time of arrival, slice duration) and the variation of slice resource demands during their life time are not considered in (Noroozi et al., 2019) and (Ebrahimi et al., 2020).

The dynamics of slice requests are considered by (Han et al., 2020) in the SAC problem. If not accepted, a request is queued for being potentially served later. The case of impatient tenants, who may leave their queues before being served, is considered. Nevertheless, neither the dynamics of resource demands within each slice, nor the activation time of a slice are accounted for. Moreover, infrastructure resources of each type are fully aggregated. As opposed to (Ebrahimi et al., 2020) and to our work, none of the details about the structure of the slice and of the infrastructure are considered in the resource model. Consequently, the proposed mechanism does not allow provision the slice in addition to admission control.

Online SAC is considered in (Bega et al., 2017) and (Bega et al., 2020) leveraging on machine learning approaches. The aim is to maximize the revenue of the InP while guaranteeing the SLAs of the admitted slices. Both papers focus on radio resources of base stations. In (Bega et al., 2017), two different types of slices are considered to account for elastic and inelastic traffic. An admissibility region is determined first, indicating the maximum number of slices that the system can support without breaking the SLAs. Both works formalize the admission control problem into a semi-Markov decision process and derive the optimal policy based on the respective request arrival parameters are known. The approach has a high computational cost and is off-line (requires system parameters to be known a priori). An alternative Q-learning approach is proposed in (Bega et al., 2017) to adapt to changing environments while achieving close to optimal performance. In (Bega et al., 2020), a deep reinforcement learning method is developed to overcome the scalability issue of the Q-learning approach. Both works consider tenants submitting slice requests for an immediate deployment, contrary to our work, where slice requests are assumed to be submitted for an immediate but also for future deployment, which permits the development of a resource provisioning strategy.

E. Slice Resource Provisioning

The topic of slice resource provisioning is relatively new to the area of network slicing and has thus a limited coverage in the literature at the time being. One may find, for instance, in (Xiong et al., 2019) and (Y. Sun et al., 2019), a preliminary study on the problem of joint resource provisioning and resource allocation for network slicing. In these papers, different slice resource provisioning frameworks in a virtualized radio access network context are introduced, where the heterogeneity of service requirements is considered. Provisioning is performed at the resource block (RB) level. The problem of radio resource provisioning and allocation from base stations (BSs) to a slice, and the assignment of users within the slices to BSs are considered. The first problem (provisioning and allocation) is solved in (Xiong et al., 2019) via heuristics, while a deep reinforcement learning technique is considered in (Y. Sun et al., 2019). The second problem (user assignment) is cast in the framework of an NP-complete 0-1 multiple knapsack problem. In these papers, the slice resource provisioning problem is studied, but is limited to radio resources.

IV. SUMMARY OF CONTRIBUTIONS

A. Network Resource Provisioning for Slicing

Related publications: (Q.-T. Luu et al., 2018) and (Kerboeuf et al., 2023).

To address Challenge 1, this thesis proposes solutions that provision resources for slices to accommodate slice resource demands. The proposed approach goes beyond previous best-effort approaches, where the SFCs of a slice are deployed sequentially in the infrastructure network. With the approach proposed in this thesis, once resources are provisioned for a given slice, the SFCs of that slice are ensured to get enough resources to operate properly. This facilitates the satisfaction of the contracted service requirements with desired quality. In addition, numerical results show that the proposed provisioning solutions yield a reduction of the computational resources needed to deploy the SFCs (see Figure 3).

In our provisioning approach, resource demands of a given slice are the aggregate resource demands of users associated to that slice. The aggregate resource demands of a slice are stated in the SLA between the MNO and the InP (MI-SLA). The MI-SLA may also include other required constraints, e.g., the successful provisioning probability when accounting for the uncertainties of slice resource demands. When performing the slice resource provisioning, the MI-SLA must be satisfied, thus guaranteeing enough infrastructure resources are provisioned for the targeted slices.

B. Coverage-Aware Slice Resource Provisioning

Related publications: (Q.-T. Luu et al., 2020) and (Q. T. Luu et al., 2020).

Challenge 2 has been addressed in Chapter 6. This chapter considers the problem of provisioning joint core and radio access network resources, accounting for some coverage constraints. To address the problem of user location (unknown during the resource provisioning phase), we have adopted a subarea partitioning approach. The coverage areas of slices are partitioned in subareas and, instead of provisioning radio blocks to users, one tries to provision radio blocks to each subarea. Several additional constraints have been presented to satisfy the coverage requirements. The main coverage constraints include: a constraint to ensure the provisioned radio resources (RBs) do not exceed the capacity of RRHs; a constraint to satisfy the minimum average user demand and the total slice radio resource demand for both uplink and downlink traffic; and finally, a constraint to ensure the proportionality between provisioned radio resources for uplink and downlink. These additional constraints lead to a complicated optimization problem when considering the problem of joint radio and network resource provisioning.

A graph of a number of different types of embedding Description automatically generated

(a) Embedding cost.

A graph with a number of numbers and a line Description automatically generated

Figure 3: Comparison of (a) embedding cost and (b) computing time between the proposed slice resource provisioning method and the traditional approach (direct embedding of slice’s SFCs).

(b) Computing time.

The joint provisioning problem (called one-step provisioning) becomes intractable when the number of slices increases. To cope with this issue, we have introduced an alternative approach (called two-step provisioning), in which the radio resource provisioning and network resource provisioning are performed sequentially. This approach is shown to have a lower time complexity than that of the joint approach, while still yielding good performance in terms of provisioning cost and efficiency of infrastructure resource utilization.

C. Uncertainty-Aware Slice Resource Provisioning

Related publication: (Q. T. Luu et al., 2021).

As pointed out in Chapter 7, the dynamics of traffics in individual slices (flow arrivals/departures), as well as of resource availability on the network infrastructure, may lead to slice QoS below the level expected by the Service Provider managing the slice. The traditional approach, in which allocated/provisioned resources are tailored to peak demands, may lead to over-provisioning of resources, thus decreasing the efficiency of infrastructure resource utilization.

In Chapter 7, a slice resource provisioning method robust to randomness of resource demands has been proposed to address Chellenge 3. The randomness is due to a partly unknown number of users with a random usage of the slice resources. The robustness is achieved by providing a probabilistic guarantee that the amount of provisioned network resources for a slice will meet the slice requirements. The proposed method also tries to maintain the impact of resource provisioning on those background services (which are also time-varying) at a prescribed level.

D. Slice Admission Control with Uncertainties

Related publications: (Q.-T. Luu et al., 2021) and (Q. T. Luu et al., 2022).

The final challenge (Challenge 4) has been addressed in Chapter 8. Slice requests are characterized by variable delays between their submission and activation time; and by different priority levels (e.g., Premium and Standard). We designed a prioritized slice admission control and resource provisioning mechanism. Admission decisions are provided and resources required for admitted slices are provisioned with a response delay depending on the slice priority and on the time left before its activation. In addition, different processing strategies have been proposed, each of which has a different impact on the processing of slice requests of different priority levels.

Numerical results show that the proportion of admitted slices can be efficiently adjusted depending on the difference in the processing delay between Premium and Standard slices. When the delay difference increases, Premium slice requests are granted significantly more frequently, with less adjustments with time in the provisioning scheme. This directly impacts the provisioning costs, which are reduced for Premium slices compared to Standard slices when the delay difference is large.

V. PERSPECTIVES

A. Accounting Additional Constraints

As discussed in Chapter 2, 5G aims to guarantee services with higher capacity, higher speed, and lower latency. To support diversified services with different requirements, some additional constraints, e.g., latency or end-to-end error rate probability, should be added to the optimization formulation. For instance, for URLLC services, which require stringent constraints in terms of latency (order of milliseconds), some latency constraints should be taken into account. In general, latency comes from several sources such as transmission delay, propagation delay, queuing, processing delays, etc. The combination of those delay sources produces a complex and variable network latency profile. Some latency constraints have been considered in the literature for the SFC embedding problem, e.g., (Alleg et al., 2017; Qu et al., 2019). In an SFC embedding problem, only one infrastructure path is used to map to one SFC. It is thus easier to formulate the latency constraints than when considering a slice resource provisioning problem, where a slice may stretch across multiple paths in the infrastructure. A way to address this issue is to perform a worst-case analysis.

B. Multi-Domain Network Slicing

One of the challenging problems in network slicing is to deploy end-to-end network slices, which refer to slices that span across multiple domains, not just a single domain. For example, network slices may stretch across a huge geographic area at a worldwide level, or encompass areas where slice coverage can only be guaranteed by using resources from different MNOs or InPs.

Similarly, some specific services may need computing and storage resources offered by a particular cloud provider (Taleb et al., 2019). In such situations, the deployment of network slices requires an efficient combination of resources provided by different InPs. The problem of slice resource provisioning in a multi-domain context thus becomes very challenging. Each InP may find its own resource provisioning solution for a part of the slice resource demands, and afterwards, there requires an efficient coordination between the InPs to eventually have a feasible provisioning solution that allows the slices to operate across the networks provided by those InPs.

To address this problem, one may introduce two algorithms, one is used by the InPs to solve the provisioning problem for the part of slice resource demands that each InP receives, and one another is used by a central entity that coordinates the InPs. This algorithm should return several possible solutions. And then, the InP coordinator runs the second algorithm with the results given by the InPs to find a feasible final solution that accounts for the total resource demands. Such approach has been considered in the literature, e.g., in (Boutigny et al., 2018; Fossati et al., 2020), but for the problem of SFC embedding. Further investigations are needed when considering this approach for the problem of slice resource provisioning.

REFERENCES

  1. 3GPP. (2020). Management and Orchestration; Concepts, Use Cases and Requirements. 3GPP TS 28.530 V17.0.0.

  2. 5G Americas. (2016). Network Slicing for 5G Networks & Services. White Paper.

  3. Alleg, A., Ahmed, T., Mosbah, M., Riggio, R., & Boutaba, R. (2017). Delay-Aware VNF Placement and Chaining based on a Flexible Resource Allocation Approach. Proc. CNSM, 1–7. https://doi.org/10.23919/CNSM.2017.8255993

  4. Barakabitze, A. A., Ahmad, A., Mijumbi, R., & Hines, A. (2020). 5G network slicing using SDN and NFV: A survey of taxonomy, architectures and future challenges. Computer Networks, 167, 106984.

  5. Basta, A., Kellerer, W., Hoffmann, M., Morper, H. J., & Hoffmann, K. (2014). Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem. ACM AllThingsCellular, 33–38. https://doi.org/10.1145/2627585.2627592

  6. Baumgartner, A., Bauschert, T., D’Andreagiovanni, F., & Reddy, V. S. (2018a). Towards Robust Network Slice Design under Correlated Demand Uncertainties. Proc. ICC, 1--7A. https://doi.org/10.1109/ICC.2018.8422618

  7. Baumgartner, A., Bauschert, T., D’Andreagiovanni, F., & Reddy, V. S. (2018b). Towards Robust Network Slice Design under Correlated Demand Uncertainties. Proc. ICC, 1-7A. https://doi.org/10.1109/ICC.2018.8422618

  8. Bauschert, T., & Reddy, V. S. (2019). Genetic Algorithms for the Network Slice Design Problem Under Uncertainty. Proc. GECCO Companion, 360–361. https://doi.org/10.1145/3319619.3322056

  9. Bega, D., Gramaglia, M., Banchs, A., Sciancalepore, V., & Costa-Perez, X. (2020). A Machine Learning Approach to 5G Infrastructure Market Optimization. IEEE Trans. Mobile Comput., 19(3), 498–512. https://doi.org/10.1109/TMC.2019.2896950

  10. Bega, D., Gramaglia, M., Banchs, A., Sciancalepore, V., & Samdanis, K. (2017). Optimising 5G Infrastructure Markets: The Business of Network Slicing. IEEE INFOCOM.

  11. Boutigny, F., Betgé-Brezetz, S., Debar, H., Blanc, G., Lavignotte, A., & Popescu, I. (2018). Multi-Provider Secure Virtual Network Embedding. Proc. IFIP International Conference on New Technologies, Mobility and Security (NTMS), 1–5. https://doi.org/10.1109/NTMS.2018.8328706

  12. Chatterjee, S., Abdel-rahman, M. J., & Mackenzie, A. B. (2018). Virtualization Framework for Cellular Networks with Downlink Rate Coverage Probability Constraints. Proc. IEEE GLOBECOM.

  13. Chatterjee, S., Member, S., & Abdel-rahman, M. J. (2020). On Optimal Orchestration of Virtualized Cellular Networks with Downlink Rate Coverage Probability Constraints. March.

  14. Chowdhury, M., Rahman, M. R., & Boutaba, R. (2012). ViNEYard: Virtual Network Embedding Algorithms. IEEE/ACM Trans. Netw., 20(1), 206–219.

  15. Cohen, R., Lewin-Eytan, L., Naor, J. S., & Raz, D. (2015). Near Optimal Placement of Virtual Network Functions. Proc. IEEE INFOCOM, 1346–1354. https://doi.org/10.1109/INFOCOM.2015.7218511

  16. Coniglio, S., Koster, A. M. C. A., & Tieves, M. (2015). Virtual Network Embedding Under Uncertainty: Exact And Heuristic Approaches. Proc. DRCN, 1–8. https://doi.org/10.1109/DRCN.2015.7148978

  17. D’Oro, S., Restuccia, F., Talamonti, A., & Melodia, T. (2019). The Slice Is Served: Enforcing Radio Access Network Slicing in Virtualized 5G Systems. IEEE Conf. Computer Commun. (IEEE INFOCOM), 442–450. https://doi.org/10.1109/INFOCOM.2019.8737481

  18. Ebrahimi, S., Zakeri, A., Akbari, B., & Mokari, N. (2020). Joint Resource and Admission Management for Slice-enabled Networks. Proc. IEEE NOMS, 1–7.

  19. Fendt, A., Mannweiler, C., Schmelz, L. C., & Bauer, B. (2019). An Efficient Model for Mobile Network Slice Embedding under Resource Uncertainty. Proc. ISWCS, 602–606. https://doi.org/10.1109/ISWCS.2019.8877372

  20. Fischer, A., Botero, J. F., Till Beck, M., De Meer, H., & Hesselbach, X. (2013). Virtual Network Embedding: A Survey. IEEE Commun. Surveys Tuts., 15(4), 1888–1906.

  21. Fossati, F., Moretti, S., Rovedakis, S., & Secci, S. (2020). Decentralization of 5G Slice Resource Allocation. Proc. IEEE/IFIP Network Operations and Management Symposium (NOMS), 1–9. https://doi.org/10.1109/NOMS47738.2020.9110391

  22. GSMA. (2018). Network Slicing: Use Case Requirements. GSMA Reference Document.

  23. Halabian, H. (2019). Distributed Resource Allocation Optimization in 5G Virtualized Networks. IEEE J. Sel. Areas Commun., 37(3), 627–642. https://doi.org/10.1109/JSAC.2019.2894305

  24. Han, B., Sciancalepore, V., Costa-Perez, X., Feng, D., & Schotten, H. D. (2020). Multiservice-Based Network Slicing Orchestration with Impatient Tenants. IEEE Trans. Wireless Commun., 19(7), 5010–5024. https://doi.org/10.1109/TWC.2020.2988644

  25. Huin, N., Jaumard, B., & Giroire, F. (2017a). Optimization of Network Service Chain Provisioning. Proc. IEEE ICC. https://doi.org/10.1109/ICC.2017.7997198

  26. Huin, N., Jaumard, B., & Giroire, F. (2017b). Optimization of Network Service Chain Provisioning. Proc. IEEE ICC. https://doi.org/10.1109/ICC.2017.7997198

  27. Huynh, N. Van, Hoang, D. T., Nguyen, D. N., & Dutkiewicz, E. (2019). Real-Time Network Slicing with Uncertain Demand : A Deep Learning Approach. Proc. IEEE ICC.

  28. Kaloxylos, A. (2018). A Survey and an Analysis of Network Slicing in 5G Networks. IEEE Communications Standards Magazine, 2(1), 20–21. https://doi.org/10.1109/MCOMSTD.2018.8334915

  29. Kerboeuf, S., Luu, Q. T., Kieffer, M., & Mouradian, A. (2023). Method and apparatus for mapping network slices onto network infrastructures with SLA guarantee (Patent U.S. Patent No. 11,570,043).

  30. Lee, Y. L., Loo, J., & Chuah, T. C. (2016). A New Network Slicing Framework for Multi-Tenant Heterogeneous Cloud Radio Access Networks. Proc. ICAEES, 414–420. https://doi.org/10.1109/ICAEES.2016.7888080

  31. Li, X., Samaka, M., Chan, A. H., Bhamare, D., Gupta, L., Guo, C., & Jain, R. (2017). Network Slicing for 5G: Challenges and Opportunities. IEEE Internet Comput., 21(5), 20–27. https://doi.org/10.1109/LANMAN.2016.7548842

  32. Li, X., Samaka, M., Chan, H. A., Bhamare, D., Gupta, L., Guo, C., & Jain, R. (2018). Network Slicing for 5G: Challenges and Opportunities. IEEE Internet Computing. https://doi.org/10.1109/MIC.2018.326150452

  33. Liang, C., & Yu, F. R. (2014). Wireless Network Virtualization: A Survey, Some Research Issues and Challenges. IEEE Commun. Surveys Tuts., 1–24. https://doi.org/10.1109/COMST.2014.2352118

  34. Liu, J., Lu, W., Zhou, F., Lu, P., & Zhu, Z. (2017). On Dynamic Service Function Chain Deployment and Readjustment. IEEE Trans. Netw. Service Manag., 14(3), 543–553.

  35. Lopez, D., & Telefonica, I. D. (2017). Service Function Chaining Use Cases in Mobile Networks. IETF Internet-Draft, 1–26.

  36. Luu, Q. T., Kerboeuf, S., & Kieffer, M. (2021). Uncertainty-Aware Resource Provisioning for Network Slicing. IEEE Transactions on Network and Service Management, 18(1), 79–93. https://doi.org/10.1109/HPSR52026.2021.9481832

  37. Luu, Q. T., Kerboeuf, S., & Kieffer, M. (2022). Admission Control and Resource Reservation for Prioritized Slice Requests With Guaranteed SLA Under Uncertainties. IEEE Transactions on Network and Service Management, 19(3), 3136–3153. https://doi.org/10.1109/TNSM.2022.3160352

  38. Luu, Q. T., Kerboeuf, S., Mouradian, A., & Kieffer, M. (2020). A Coverage-Aware Resource Provisioning Method for Network Slicing. IEEE/ACM Transactions on Networking, 28(6), 2393–2406. https://doi.org/10.1109/TNET.2020.3019098

  39. Luu, Q.-T. (2021). Dynamic Control and Optimization of Wireless Virtual Networks. Université Paris-Saclay.

  40. Luu, Q.-T., Kerboeuf, S., & Kieffer, M. (2021). Foresighted resource provisioning for network slicing. 2021 IEEE 22nd International Conference on High Performance Switching and Routing (HPSR), 1–8.

  41. Luu, Q.-T., Kerboeuf, S., Mouradian, A., & Kieffer, M. (2020). Radio Resource Provisioning for Network Slicing with Coverage Constraints. IEEE International Conference on Communications (ICC), 1–7.

  42. Luu, Q.-T., Kieffer, M., Mouradian, A., & Kerboeuf, S. (2018). Aggregated Resource Provisioning for Network Slices. IEEE Global Communications Conference (GLOBECOM).

  43. Mireslami, S., Rakai, L., Wang, M., & Far, B. H. (2019). Dynamic Cloud Resource Allocation Considering Demand Uncertainty. IEEE Trans. on Cloud Comput., 7161(c), 1. https://doi.org/10.1109/tcc.2019.2897304

  44. Noroozi, K., Karimzadeh-Farshbafan, M., & Shah-Mansouri, V. (2019). Service Admission Control for 5G Mobile Networks with RAN and Core Slicing. Proc. IEEE GLOBECOM, 6–11. https://doi.org/10.1109/GLOBECOM38437.2019.9013617

  45. Qu, K., Zhuang, W., Ye, Q., Shen, X. S., Li, X., & Rao, J. (2019). Delay-Aware Flow Migration for Embedded Services in 5G Core Networks. Proc. ICC, 1–6.

  46. Richart, M., Baliosian, J., Serrat, J., & Gorricho, J. L. (2016). Resource Slicing in Virtual Wireless Networks: A Survey. IEEE Trans. Netw. Service Manag., 13(3), 462–476. https://doi.org/10.1109/TNSM.2016.2597295

  47. Riera, J. F., Batall, J., Liberati, F., Giuseppi, A., Pietrabissa, A., Ceselli, A., Petrini, A., Trubian, M., Papadimitrou, P., Dietrich, D., Ramos, A., & Meli, J. (2016). TeNOR: Steps Towards an Orchestration Platform for Multi-PoP NFV Deployment. Proc. IEEE NetSoft, 243–250.

  48. Riggio, R., Bradai, A., Harutyunyan, D., & Rasheed, T. (2016). Scheduling Wireless Virtual Networks Functions. IEEE Transactions on Network and Service Management, 13(2), 240–252.

  49. Rost, P., Mannweiler, C., Michalopoulos, Di. S., Sartori, C., Sciancalepore, V., Sastry, N., Holland, O., Tayade, S., Han, B., Bega, D., Aziz, D., & Bakker, H. (2017). Network Slicing to Enable Scalability and Flexibility in 5G Mobile Networks. IEEE Commun. Mag., 55(5), 72–79. https://doi.org/10.1109/MCOM.2017.1600920

  50. Su, R., Zhang, D., Venkatesan, R., Gong, Z., Li, C., Ding, F., Jiang, F., & Zhu, Z. (2019). Resource Allocation for Network Slicing in 5G Telecommunication Networks: A Survey of Principles and Models. IEEE Netw., 33(6), 172–179. https://doi.org/10.1109/mnet.2019.1900024

  51. Sun, G., Xiong, K., Boateng, G. O., Ayepah-Mensah, D., Liu, G., & Jiang, W. (2019). Autonomous Resource Provisioning and Resource Customization for Mixed Traffics in Virtualized Radio Access Network. IEEE Systems Journal, 13(3), 2454–2465. https://doi.org/10.1109/JSYST.2019.2918005

  52. Sun, Y., Wang, Y., Yu, H., Guo, B., & Zhang, X. (2019). A learning-based bandwidth resource allocation method in sliced 5G C-RAN. 2019 IEEE Globecom Workshops, GC Wkshps 2019 - Proceedings. https://doi.org/10.1109/GCWkshps45667.2019.9024681

  53. Taleb, T., Afolabi, I., Samdanis, K., & Yousaf, F. Z. (2019). On Multi-Domain Network Slicing Orchestration Architecture and Federated Resource Control. IEEE Network, 33(5), 242–252. https://doi.org/10.1109/MNET.2018.1800267

  54. Teague, K., Abdel-Rahman, M. J., & Mackenzie, A. B. (2019). Joint Base Station Selection and Adaptive Slicing in Virtualized Wireless Networks: A Stochastic Optimization Framework. Proc. International Conference on Computing, Networking and Communications, (ICNC), 859–863. https://doi.org/10.1109/ICCNC.2019.8685518

  55. Trinh, T., Esaki, H., & Aswakul, C. (2011). Quality of Service Using Careful Overbooking for Optimal Virtual Network Resource Allocation. Proc. ECTI, 296–299. https://doi.org/10.1109/ECTICON.2011.5947831

  56. Vizarreta, P., Condoluci, M., Machuca, C. M., Mahmoodi, T., & Kellerer, W. (2017a). QoS-driven Function Placement Reducing Expenditures in NFV Deployments.

  57. Vizarreta, P., Condoluci, M., Machuca, C. M., Mahmoodi, T., & Kellerer, W. (2017b). QoS-driven Function Placement Reducing Expenditures in NFV Deployments. Proc. IEEE ICC. https://doi.org/10.1109/ICC.2017.7996513

  58. Vizarreta, P., Condoluci, M., Machuca, C. M., Mahmoodi, T., & Kellerer, W. (2017c). QoS-driven Function Placement Reducing Expenditures in NFV Deployments. Proc. IEEE ICC. https://doi.org/10.1109/ICC.2017.7996513

  59. Wang, G., Feng, G., Quek, T. Q. S., Qin, S., Wen, R., & Tan, W. (2019). Reconfiguration in Network Slicing-Optimizing the Profit and Performance. IEEE Trans. Netw. Service Manag., 16(2), 591–605. https://doi.org/10.1109/tnsm.2019.2899609

  60. Wang, G., Feng, G., Tan, W., Qin, S., Ruihan, W., & Sun, S. (2017). Resource Allocation for Network Slices in 5G with Network Resource Pricing. Proc. IEEE GLOBECOM, 1–6.

  61. Wang, Y., Li, Z., Xie, G., & Salamatian, K. (2017). Enabling automatic composition and verification of service function chain. 2017 IEEE/ACM 25th International Symposium on Quality of Service, IWQoS 2017. https://doi.org/10.1109/IWQoS.2017.7969147

  62. Weldon, M. K. (2015). The Future X Network: A Bell Labs Perspective. CRC Press.

  63. Wen, R., Feng, G., Tang, J., Quek, T. Q. S., Wang, G., Tan, W., & Qin, S. (2019). On Robustness of Network Slicing for Next-Generation Mobile Networks. IEEE Trans. Commun., 67(1), 430–444. https://doi.org/10.1109/TCOMM.2018.2868652

Xiong, W., Bogdanov, P., & Zheleva, M. (2019). Robust and Efficient Modulation Recognition Based on Local Sequential {IQ} Features. IEEE Conference on Computer Communications, 1612–1620.