What is the role of NFV in the development of mobile core networks?

the need for more flexibility and elasticity has led to the consideration of SDN and NFV as the key enablers for more dynamic EPC networks, paving the way to a network of capabilities.

What is the catalyst for enabling such a scalable multi-service mobile backhaul?

The catalyst for enabling such scalable multi-service mobile backhaul is Multi-Protocol Label Switching (MPLS) that supported the progressive adoption of different transport layer technologies unifying 2G Time Division Multiplexing (TDM) and High-level Data Link Control (HDLC) transport, 3G Asynchronous Transfer Mode (ATM) and Frame Relay as well as 4G Ethernet and IP [148] [149].

What is the definition of UE slicing?

UE slicing considers the smart devices as a commodity hardware platform having a pre-installed middleware (similar to a hypervisor on general-purpose servers or PCs), which can accommodate, manage and schedule resources between multiple mobile OS entities.

What are the main challenges of network slicing?

Although network slicing is currently undergoing a standardization phase, there are still numerous open research problems and implementation challenges to be addressed.

What is the advantage of a service-oriented NG core architecture?

The service-oriented NG core architecture could be easily deployed in a software-based/virtualized environment (e.g. VM or container), wherein libraries of functions may be requested from a VNF catalog and composed into end-toend service chains on demand.

What is the definition of a network slice orchestration architecture?

The network slice orchestration architecture consists of:• End-to-end service management and orchestrator: receives network slice requests from verticals and third parties, and creates a slice by performing slice brokering, admission control, policy provisioning and service mapping, considering the desired SLA, customization and slice template.

What is the main advantage of the functional split?

This flexible functional split can highly impact the performance of network slicing and the optimal split largely depends on the characteristics of the target service.

What is the concept of core network slicing?

the notion of core network slicing is centered around the possibility to deploy multiple instances of virtual EPC (vEPC) running in parallel in order for each to fulfill different service demands, e.g. a delay-sensitive service may require a distributed vEPC closer to the end user.

What is the role of a RAN centralization in the mobile backhaul?

Initial efforts to address such requirements considered: (i) small cell enhancements for the mobile backhaul focusing on scalable connectivity and various coordination types with the macro, i.e. tight or loose [151], and (ii) RAN centralization using the Common Public Radio Interface (CPRI) interface [152], which allows an ideal optical fiber fronthaul.

What is the role of the MOP in the network sharing management paradigm?

This network sharing management paradigm introduces the Master Operator (MOP) as a single actor, which is responsible for the shared infrastructure deployment, including the spectrum and related operations.

What is the way to ensure isolation and performance guarantees between the different logical networks?

Network slicing can assure isolation and performance guarantees between the different logical networks that employ a different fronthaul/backhaul flavor according to the corresponding base station functional split.

Why does a slice tenant have the highest level of control and privileges?

This is because the SLA allows the slice consumer to operate and manage his/her own virtualization platform and other related management support systems, and hence can deploy any form of network functions as needed to deliver the desired set of network services.

What is the advantage of connecting distributed RRH physical layers to a common MAC?

This option leverages the benefits of connecting distributed RRH physical layers to a common MAC, which allows coordinated scheduling and dynamic point selection, i.e. Coordinated Multi-Point (CoMP).

How much of the cost of investment could be saved by network slicing?

Such savings on the investment cost could even rise to as high as more than half of the total cost of investment, when adopting active resource brokering in the form of network slicing.

Why is the network slice resource allocation problem so complex?

This is due to the fact that network slices, especially those which are not deployed for a known number of users, often face the challenge of insufficient resources as the number of users increases, thereby resulting into poor network performance.

What is the advantage of virtualizing the RLC layer and other layers above it?

The RLC layer and other layers above it are virtualized at the BBU allowing multiple MAC entities to be associated with a common RLC entity.

(Open Access) Network Slicing and Softwarization: A Survey on Principles, Enabling Technologies, and Solutions (2018) | Ibrahim Afolabi

Q: What have the authors contributed in "Network slicing & softwarization: a survey on principles, enabling technologies & solutions" ?

This paper elaborates network slicing from an endto-end perspective detailing its historical heritage, principal concepts, enabling technologies and solutions as well as the current standardization efforts. This paper also provides details of specific slicing solutions for each part of the 5G system. Finally, this paper identifies a number of open research challenges and provides recommendations towards potential solutions.

This is an electronic reprint of the original article.

This reprint may differ from the original in pagination and typographic detail.

This material is protected by copyright and other intellectual property rights, and duplication or sale of all or

part of any of the repository collections is not permitted, except that material may be duplicated by you for

your research use or educational purposes in electronic or print form. You must obtain permission for any

other use. Electronic or print copies may not be offered, whether for sale or otherwise to anyone who is not

an authorised user.

Afolabi, Ibrahim; Taleb, Tarik; Samdanis, Konstantinos; Ksentini, Adlen; Flinck, Hannu

Network Slicing & Softwarization

Published in:

IEEE Communications Surveys and Tutorials

DOI:

10.1109/COMST.2018.2815638

Published: 01/01/2018

Document Version

Peer reviewed version

Please cite the original version:

Afolabi, I., Taleb, T., Samdanis, K., Ksentini, A., & Flinck, H. (2018). Network Slicing & Softwarization: A Survey

on Principles, Enabling Technologies & Solutions. IEEE Communications Surveys and Tutorials, 20(3), 2429 -

2453. https://doi.org/10.1109/COMST.2018.2815638

This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.

The final version of record is available at  http://dx.doi.org/10.1109/COMST.2018.2815638

Network Slicing & Softwarization: A Survey

Principles, Enabling Technologies & Solutions

Ibrahim Afolabi, Tarik Taleb, Konstantinos Samdanis, Adlen Ksentini, and Hannu Flinck

Abstract—Network slicing has been identiﬁed as the backbone

of the rapidly evolving 5G technology. However, as its consolida-

tion and standardization progress, there are no literatures that

comprehensively discuss its key principles, enablers and research

challenges. This paper elaborates network slicing from an end-

to-end perspective detailing its historical heritage, principal

concepts, enabling technologies and solutions as well as the

current standardization efforts. In particular, it overviews the

diverse use cases and network requirements of network slicing,

the pre-slicing era, considering RAN sharing as well as the end-

to-end orchestration and management, encompassing the radio

access, transport network and the core network. This paper

also provides details of speciﬁc slicing solutions for each part

of the 5G system. Finally, this paper identiﬁes a number of

open research challenges and provides recommendations towards

potential solutions.

Index Terms—Network Slice, 5G, Network Softwarization,

Orchestration, Network Management, NFV, SDN, Cloud, Mobile

Network, MANO and Open Source.

I. INTRODUCTION

The emerging 5G mobile system is expected to build on

the success of the current 4G technology offering support

for a plethora of network services with diverse performance

requirements. 5G era is touted as the generation of mobile

networks that will support dedicated use-cases and provide

speciﬁc types of services to satisfy simultaneously various

customer demands. Unlike the “one-ﬁt-all” type of the 4G

architecture, 5G is anticipated to consider diverse business

demands with often conﬂicting requirements encouraging

service innovation and programmability through the use of

open sources and open interfaces that allow access to third

parties. By allowing different parties to instantiate and run a

software-based architecture, 5G becomes inherently a multi-

tenant ecosystem, whereby a tenant refers to a user or group of

users with speciﬁc access rights and privileges over a shared

resource. Hence, 5G networks offer multi-tenancy support and

service-tailored connectivity, providing a top-notch Quality

of Service (QoS) which will ultimately result in a long

lasting Quality of Experience (QoE) with a truly differentiated

service provisioning on top of a shared underlying network

infrastructure.

5G networks are also expected to create new service ca-

pabilities relying on recent advancements in the Internet of

Things (IoT) area. In particular, analysts forecast that by 2025

Ibrahim Afolabi is with Aalto University, Espoo, Finland. Tarik

Taleb is also with Aalto University, Espoo, Finland and Sejong Uni-

versity, Seoul, Korea. Konstantinos Samdanis is with Huawei European

Research Center, Munich, Germany. Adlen Ksentini is with Eurecom,

Nice, France. Hannu Flinck is with Nokia Bell Labs, Espoo, Finland.

Emails:{ﬁrstname.lastname}@aalto.ﬁ, konstantinos.samdanis@huawei.com,

adlen.ksentini@eurecom.fr, hannu.ﬂinck@nokia-bell-labs.com.

the number of IoT devices could grow to a stunning ﬁgure of

about 100 billion devices [1], supporting a wide range of ser-

vices spanning from low-cost sensor-based metering services

and delay-tolerant vehicle services to critical communications

including e-health, e-business and automotive. For mobile

operators, IoT does not mean only support for many more

devices and massive connectivity, but also deﬁnes a promising

opportunity for offering novel services and business solutions

within the IoT value chain beyond simple connectivity. To

this end, 5G enables open interfaces to support vertical seg-

ments, i.e. third parties not owning network infrastructure and

requiring networking services with speciﬁc needs, as well as

new business solutions, e.g., AT&T digital life customized to

the needs of users. The automotive industry deﬁnes one of

the most signiﬁcant 5G vertical segments. It requires efﬁcient

networking capabilities combined with IoT and edge-cloud to

facilitate a number of services including autonomous driving,

bird eye view, and real-time assessment of road conditions,

just to mention a few.

5G should leverage the beneﬁts of network virtualization

to accommodate ﬂexibility in providing carrier-grade differ-

entiated mobile network services and ubiquitous coverage,

unifying heterogeneous radio and networking technologies

supporting the existing 3G (3rd Generation), LTE (Long

Time Evolution) and Wi-Fi technologies, and efﬁciently inter-

working them with the emerging 5G new radio and ﬁxed

access networks. The notion of network virtualization concen-

trates on the concept of a software-based representation of both

the hardware and software resources considering both data

and/or control-plane functions. It is the main foundation of (i)

network softwarization and (ii) network slicing. Network soft-

warization

is the concept of designing, architecting, deploy-

ing and managing network components, primarily based on

software programmability properties [2]. It enables ﬂexibility,

adaptability, and even total reconﬁguration of a network on the

ﬂy based on timely requirements and behaviors by considering

cost and process optimization in the overall maintenance of

the network lifecycle. Network slicing on the other hand

seeks to assure service customization, isolation and multi-

tenancy support on a common physical network infrastructure

by enabling logical as well as physical separation of network

resources.

Softwarization is expected to impact several aspects of net-

work development and services such as Content Delivery Net-

works (CDN) or video accelerators [3] [4]. It has shown huge

potential in revolutionizing the deployment and operations of

mobile networks, by simply untying network functions from

Softwarization encompasses orchestration.

proprietary hardware and enabling them to run on commercial

off-the-shelf (COTS) hardware computers or data-centers. This

possibility makes it doable to deploy software programs as

feature updates/upgrades to the necessary network parts to

enable newer network functions or simply ﬁxing bugs in

existing ones. The feasibility to plug newer functions into the

network through programmable interfaces makes the network

more ﬂexible, scalable, elastic and perhaps reactive through

the use of technologies such as artiﬁcial intelligence [5].

Network slicing has been recently gaining momentum

among an ever-growing community of researchers from both

academia and industry. It has been also the focus of different

standardization bodies (e.g., 3GPP (3rd Generation Partnership

Project), IETF (Internet Engineering Task Force) and ITU-T

(International Telecommunication Union - Telecommunication

Standardization Sector)). This concept can be traced back to

the idea of Infrastructure as a Service (IaaS) cloud computing

model, whereby different tenants share computing, networking

and storage resources, in order to create different isolated

fully-functional virtual networks on a common infrastructure.

In the context of 5G and beyond, a network slice is a uniﬁca-

tion of virtual resources (e.g., VMs) wherein a set of Virtual

Network Functions (VNF) are instantiated and connected via

a virtual network, e.g., Virtual Local Area Network (VLAN)

and Virtual Private Network (VPN). The possibility to create,

on demand and in a programmable fashion, cost-efﬁcient end-

to-end network slices and dedicate them for the dynamic

provisioning of diverse services is seen as an important feature

of 5G. In this vein, efforts are ongoing towards developing

a 5G mobile system capable of deploying network slices of

varying sizes and structures.

To the best knowledge of the authors, there are no detailed

surveys on the topic of network slicing in the literature, except

those presented in [6], [7], [9]. In [6], the authors presented

a brief analysis of the state of the art of network slicing in

5G using a framework to evaluate the maturity state of the

identiﬁed projects and their relevance towards the 5G network

evolution. Likewise, in [7], the authors describe how wireless

network resources can be adequately allocated to network

slices without the resources of one slice negatively impacting

the quality of others and with a focus on the Radio Access Net-

work (RAN). In [9], the authors study the notion of network

slicing in 5G following an architecture model that consists

of the infrastructure layer, network function layer and service

layer providing an overview and the corresponding challenges.

In [8], the authors focus on how Multi-access Edge Computing

(MEC) is advantageous to the RAN, especially with respect to

latency, its enabling technologies, and orchestration options.

Other relevant articles in light of network slicing and its

enablers such as Network Functions Virtualization (NFV) and

Software Deﬁned Networking (SDN) are also presented in

[10], [11], [85] and [12]–[14], respectively.

In comparison to these surveys, the contributions of this

paper are manifold. First, this paper presents an extensive

and exhaustive review of the principal concepts and enabling

technologies for facilitating end-to-end network slicing encom-

passing all aspects of the 5G and networking technologies

from the access and core networks, to the transport networks

in a holistic manner. It also overviews the key business drivers

and major ongoing research projects in line with the automa-

tion and orchestration of end-to-end network slices as well

as its lifecycle management. This survey also delves into the

roots of network slicing as well as its emerging technologies

detailing their various impacts in the architectural evolution

of 5G networks with particular focus on slice orchestration

and management. Finally, we identify and discuss a number

of open research challenges relevant to security [15] as well

as network slice resource allocation.

The rest of this paper is organized in the following fashion.

Section II presents the 5G service and business requirements,

focusing mainly on 5G service requirements and business

drivers for emerging markets as enumerated by IMT-2020

(International Mobile Telecommunication system - beyond

2020) and the European Commission (EC) 5G Infrastructure

Public Private Partnership (5GPPP). Section III provides an

overview of the network slicing concept and presents its

main use cases with a glimpse into pre-5G network slicing.

Section IV provides an overview of 3GPP network sharing,

highlighting how it started, discussing its relevance to network

slicing, and presenting its variants. As enablers of network

slicing, different virtualization technologies are detailed in

Section V. Section VI describes network slice orchestration

and management, with particular focus on network slice or-

chestration architecture, broker, capacity provisioning policy,

lifecycle management and explains how network slices can

be federated across multiple administrative domains. Section

VII analyzes RAN slicing, explaining the RAN slicing re-

quirements, slice resource management and isolation, RAN

programmability, and RAN functional split as well as the

fronthaul/backhaul slice transport options. Section VIII sheds

light on core network slicing, discussing separately the slicing

principles of the Evolved Packet Core (EPC) and the new

5G core network. Section IX elaborates pending research

problems and challenges relevant to network slicing. The paper

concludes in Section X.

II. 5G SERVICE & BUSINESS REQUIREMENTS

A. 5G Service Requirements

5G networks are anticipated to revolutionize the user experi-

ence introducing new requirements to shape network platforms

for launching new innovative services. These services have

diverse requirements, involving higher data trafﬁc volumes

and potential number of devices. The initial roll out of 5G

is expected by 2020 in order to meet the emerging business

and consumer demands. The IMT-2020 vision assists the

development of various industry sectors [16], introducing the

following targets for research and innovation:

• low latency, i.e. 1ms over-the-air, and high reliability,

• user density with area trafﬁc capacity of 10 Mbps/m

• peak data rate of 10 Gbps with particular scenarios

supporting up to 20 Gbps,

• service continuity under high mobility with 500km/h,

• connection density with 10

devices per km

• 100 Mbps user experienced data rates for wide area

coverage,

This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.

The final version of record is available at  http://dx.doi.org/10.1109/COMST.2018.2815638

• three times higher spectral efﬁciency (i.e., in comparison

to 4G),

• 100 times more energy-efﬁcient networking, and

• energy lifetime for sensors to be greater than 10 years.

5GPPP [17] that encourages research towards 5G, brought

light into the requirements of 5G through the ﬂagship project

METIS introducing the following target network capabilities

[18]:

• Amazingly fast: A feature that shall enable instantaneous

network connectivity for all applications by providing

10Gbps data rates.

• Great service in a crowd: A feature that shall enable

a broadband experience, regardless of the user density,

assuring a trafﬁc volume of 9 Gbytes/h and a data rate

up to 20 Mbps per user.

• Best experience follows you: A feature that allows a ﬁxed

line network experience for users on the move, with at

least 100 Mbps in the downlink and 20 Mbps in the

uplink.

• Super real-time and reliable connection: A feature that

aims at supporting mission-critical machine type commu-

nications, ensuring 99.999% reliability and less than 5ms

end-to-end latency.

• Ubiquitous things communicating: A feature that aims at

providing wireless connectivity for sensors and actuators,

supporting 300,000 devices per cell, while prolonging the

battery lifetime of devices in the order of a decade.

These main service requirements are encouraging rapid

time-to-market for launching new services (e.g., deployment

time shorter than 90 minutes) and at reducing network man-

agement Operational Expenditures (OPEX) by 80%. IMT-2020

and 5GPPP have identiﬁed the need for enhanced security and

privacy, without explicitly quantifying them.

B. Business Drivers & Emerging Markets

5G is expected to facilitate a business ecosystem, enabling

innovative services and networking capabilities not only for

consumers, but also for new industry stakeholders. Hence,

5G needs to adopt new partnerships and business models for

different types of customers, being the key asset for enabling

vertical industries and contributing to the fourth industrial rev-

olution impacting multiple sectors [19]. Verticals can facilitate

the development of new products and services, while network

operators can create partnerships to accelerate network service

roll-outs or to create customized services to vertical industries.

The business roles that the 5G architecture would facilitate

through virtualization and slicing are the following:

• Infrastructure providers: offer the physical network in-

frastructure and are responsible for upgrades and main-

tenance. Currently, network operators take the role of

the infrastructure providers. However, in the emerging

5G, third players can provide networking hardware and

connectivity in private or community indoor areas, e.g.,

in a stadium or shopping mall.

• Cloud providers: facilitate third parties with computa-

tion and storage resources and potential cloud services,

e.g., platform-as-a-service such as Linux’s Openstack,

Amazon web service’s Elastic Compute Cloud (EC2),

Google’s Kubernetes, and Microsoft’s Azure.

• Virtual network operators: lease resources from an infras-

tructure provider to either complement their own capacity

and/or coverage (e.g., Lycabmobile, Lebara, and Virgin

Mobile), or gain network coverage in case they lack

physical infrastructure. Such leased resources can help

against complex and lengthy processes for site acquisition

in urban areas as well as to enhance network coverage

with low risk in remote areas.

• Service broker: interacts with the physical network, col-

lects abstracted resource information and acts as a medi-

ator mapping the service requests originated from virtual

network operators, application providers and verticals, to

the mobile network operator’s resources. A service broker

can be a component of the infrastructure provider, mobile

network operator or an independent third party.

• Application providers: offer, with best-effort perfor-

mance, services operating on top of a network belonging

to an operator. 5G applications with a high data con-

sumption may push application providers (e.g., Netﬂix

and Hulu), to buy network resources from operators, in

order to encourage end-users to consume their services

without being charged per data volume usage. In addition,

application with stringent requirements may pre-deﬁne a

Service Level Agreements (SLA) set of requirements with

operators to ensure a satisfying user experience.

• Verticals: offer a variety of services to a non-telecom

speciﬁc industry, exploiting network and cloud resources

from network operators and cloud providers. Most of

the new growth is anticipated in taking place through

digitalization of the vertical industries such as factories,

transportation and health care.

Partnerships between different business players can be

established over networking and cloud resources, network

capabilities exposure, value-added services and network con-

text information as well as on providing 5G services as a

programmable and software oriented capability set. Network

slicing is a key technology and business enabler for 5G,

facilitating multi-tenancy and enhanced network coverage for

third parties in a ﬂexible way, assuring an extra revenue means

for operators, infrastructure and cloud providers. Network

slices can be established on a permanent basis or on-demand,

either opportunistically or periodically, with network and cloud

resources belonging to a single or multiple operators or to a

mix of different business players.

III. NETWORK SLICING CONCEPT & USE-CASES

A. Early pre-5G Network Slicing

Network slicing rely on virtualization concepts, which have

been around as far back as the 1960s [20] [21] when the

ﬁrst operating system (CP-40) was developed by IBM [22].

The design of the CP-40 on IBM system 360/40 supported

time-sharing and virtual memory, introducing a breakthrough

in computing by accommodating up to ﬁfteen users simulta-

neously [23] with the illusion of working individually on a

complete set of hardware and software [20] [23]. The idea of

This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.

The final version of record is available at  http://dx.doi.org/10.1109/COMST.2018.2815638

virtualization, i.e. creating a virtual form of a physical entity

through software methods and processes, formed the vision of

virtual systems spanning across computing platforms, network

resources, and storage devices [22]. Virtualization was widely

adopted for data centers in the 70s and by the early 80s, it was

applied into networking, for connecting remote sites securely

with controlled performance through the Internet.

The introduction of overlay networks in the late 80s that

consist of nodes connected over logical links forming a virtual

network over a network composed of physical infrastructure

can be seen as an early form of network slicing, com-

bining heterogeneous resources over various administrative

domains. Overlay networks provide QoS guarantees in a

service-oriented fashion. They are ﬂexible in nature but not

automated nor programmable. By 2000, the ﬁrst-generation

platforms for verifying and evaluating new network protocols

were established based on overlay networks. PlanetLab [24]

[25] adopted a common software package called MyPLC

enabling distributed virtualization by allowing users to obtain

isolated application speciﬁc slices. A slice was deﬁned as a

unit component with allocated resources such as computing

power/storage on servers or resources existing in namespaces.

However, such overlay platforms had limitations in underlay

network controls.

In 2008, the GENI project, a US National Science Foun-

dation (NSF) [26] initiative, pushed forward the development

of a testbed based on network virtualization technologies for

promoting research on a clean slate network, while considering

federated resources and mobile network environments [27].

GENI offers instrumentation and measurement tools used for

carrying out both active and passive measurements and for

visualizing and analyzing measurement results [28]. By 2009,

Software Deﬁned Network (SDN) technologies enabled re-

searchers to run their experiments in a slice of existing campus

networks allowing programmability via open interfaces [29].

B. The 5G Network Slice Concept & Principles

Network slicing in the context of 5G is a new deﬁned

concept introduced by NGMN (Next Generation Mobile Net-

work) in [30]. Network slicing facilitates multiple logical

self-contained networks on top of a common physical in-

frastructure platform enabling a ﬂexible stakeholder ecosys-

tem that allows technical and business innovation integrating

physical and/or logical network and cloud resources into a

programmable, open software-oriented multi-tenant network

environment. 3GPP deﬁnes network slicing as a technology

that “enables the operator to create networks, customized to

provide optimized solutions for different market scenarios

which demand diverse requirements, e.g. in terms of func-

tionality, performance and isolation [31]. For ITU-T, network

slicing is perceived as Logical Isolated Network Partitions

(LINP) composed of multiple virtual resources, isolated and

equipped with a programmable control and data plane [32].

Network slicing enables value creation for vertical seg-

ments, application providers and third parties that lack physical

network infrastructure, by offering radio, networking and

cloud resources, allowing a customized network operation and

true service differentiation. The VNFs, which constitute a

network slice, may vary drastically depending on the service

requirements of that particular slice. The type of service

associated with a network slice would determine the resources

and service treatment the network slice would receive, e.g.

a real-time communication network slice would receive the

appropriate resources and service treatment to meet ultra low

latency demands [33]. Network slicing builds on top of the

following seven main principles that shape the concept and

related operations:

• Automation: enables an on-demand conﬁguration of net-

work slicing without the need of ﬁxed contractual agree-

ments and manual intervention. Such convenient opera-

tion relies on signaling-based mechanisms, which allow

third parties to place a slice creation request indicating

besides the conventional SLA which would reﬂect the

desired capacity, latency, jitter, etc., timing information

considering the starting and ending time, and duration or

periodicity of a network slice.

• Isolation: is a fundamental property of network slic-

ing that assures performance guarantees and security

(to defend network openness to third parties) for each

tenant even when different tenants use network slices

for services with conﬂicting performance requirements.

However, isolation may come at the cost of reducing

multiplexing gain, depending on the means of resource

separation for explicit use, which may result in inefﬁcient

network resource utilization. The notion of isolation

involves not only the data plane but also the control

plane, while its implementation deﬁnes the degree of

resource separation. Isolation can be deployed (i) by using

a different physical resource, (ii) when separating via

virtualization means a shared resource and (iii) through

sharing a resource with the guidance of a respective

policy that deﬁnes the access rights for each tenant.

• Customization: assures that the resources allocated to a

particular tenant are efﬁciently utilized in order to meet

best the respective service requirements. Slice customiza-

tion can be realized (i) in a network wide level consider-

ing the abstracted topology and the separation of data and

control plane, (ii) on the data plane with service-tailored

network functions and data forwarding mechanism, (iii)

on the control plane introducing programmable policies,

operations and protocols and (iv) through value-added

services such as big data and context awareness.

• Elasticity: is an essential operation related with the re-

source allocated to a particular network slice, in order

to assure the desired SLA under varying (i) radio and

network conditions, (ii) amount of serving users, or (iii)

geographical serving area because of user mobility. Such

resource elasticity can be realized by reshaping the use of

the allocated resources by scaling up/down or relocating

VNFs and value-added services, or by adjusting the

applied policy and re-programing the functionality of

certain data and control plane elements. Elasticity can

also take the form of altering the amount of initially

allocated resources by modifying physical and virtual

This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.

The final version of record is available at  http://dx.doi.org/10.1109/COMST.2018.2815638