Edinburgh Research: 5G Network Slicing Survey and Challenges

This assignment is a comprehensive survey on 5G network slicing, a critical technology for enabling diverse services with varying performance requirements. The document begins by introducing the concept of network slicing, highlighting its importance in the context of 5G's evolution to support a wide range of applications. It then delves into the architectural aspects, discussing representative 5G architectural proposals and presenting a generic framework that encompasses infrastructure, network function, and service layers, along with service management and orchestration (MANO). The core of the assignment reviews existing work on 5G network slicing, evaluating the maturity of current proposals within the defined framework. The survey identifies key challenges that must be addressed to realize a softwarized 5G mobile network architecture, including the virtualization of radio resources, the composition of network infrastructure, and the need for efficient resource allocation across different network slices. The assignment concludes by emphasizing the need for further research and development to overcome these challenges and achieve the full potential of 5G network slicing.

Edinburgh Research Explorer
Network Slicing in 5G: Survey and Challenges
Citation for published version:
Foukas, X, Elmokashfi, A, Patounas, G & Marina, MK 2017, 'Network Slicing in 5G: Survey and Challenges'
IEEE Communications Magazine, vol. 55, no. 5, pp. 94-100. https://doi.org/10.1109/MCOM.2017.1600951
Digital Object Identifier (DOI):
10.1109/MCOM.2017.1600951
Link:
Link to publication record in Edinburgh Research Explorer
Document Version:
Peer reviewed version
Published In:
IEEE Communications Magazine
General rights
Copyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s)
and / or other copyright owners and it is a condition of accessing these publications that users recognise and
abide by the legal requirements associated with these rights.
Take down policy
The University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorer
content complies with UK legislation. If you believe that the public display of this file breaches copyright please
contact [e-mail] providing details, and we will remove access to the work immediately and
investigate your claim.

1
Network Slicing in 5G: Survey and Challenges
Xenofon Foukas∗
, Georgios Patounas†, Ahmed Elmokashfi†, Mahesh K. Marina∗
∗
The University of Edinburgh, Edinburgh, United Kingdom
†Simula Research Laboratory, Oslo, Norway
∗
{x.foukas, mahesh}@ed.ac.uk, †{gpatounas, ahmed}@simula.no
Abstract—5G is envisioned to be a multi-service network
supporting a wide range of verticals with a diverse set of
performance and service requirements. Slicing a single physical
network into multiple isolated logical networks has emerged as
a key to realizing this vision. This article is meant to act as
a survey, the first to the authors’ knowledge, on this topic of
prime interest. We begin by reviewing the state of the art on
5G network slicing and we present a framework for bringing
together and discussing existing work in a holistic manner. Using
this framework, we evaluate the maturity of current proposals
and identify a number of open research questions.
I. INTRODUCTION
Mobile devices have become an essential part of our daily
lives and as such the mobile network infrastructure that
connects them has become critical. It is set to take on an
even bigger role with the next-generation 5G mobile systems
envisioned to support a wide array of services and devices.
In this article, we consider the architectural aspect of mobile
networks looking towards 5G. Examining the evolution of
mobile networks until now suggests that the changes across
generations have been driven largely by the need to support
faster data oriented services. For instance, spectral efficiency
in the Radio Access Network (RAN) has increased by a factor
of 30 from 2G to 4G. On the Core Network (CN) front, the
packet switched (IP) component introduced initially in the
2.5G (GPRS) system eventually supplanted the legacy circuit
switched component in 4G systems.
What 5G systems are going to be is yet to be determined.
However, it is conceivable that the eventual 5G system would
be a convergence of two complementary views that are cur-
rently driving the research and industrial activity on 5G. One
is an evolutionary view focusing on significantly scaling up
and improving the efficiency of mobile networks (e.g., 1000x
traffic volume, 100x devices, and 100x throughput). Much of
the research focus around this view is on the radio access
side looking at novel technologies and spectrum bands (e.g.,
massive MIMO, millimeter waves).
The other service-oriented view envisions 5G systems to
cater to a wide range of services differing in their requirements
and types of devices, and going beyond the traditional human-
type communications to include various types of machine-type
communications. This requires the network to take different
forms depending on the service in question, leading naturally
to the notion of slicing the network on a per-service basis,
the focus of this article. Realizing this service-oriented view
requires a radical rethink of the mobile network architecture
to turn it into a more flexible and programmable fabric,
leveraging technologies like SDN and NFV, that can be used
to simultaneously provide a multitude of diverse services over
a common underlying physical infrastructure. We take this
view in this article as it is intertwined with the 5G mobile
network architecture, although the evolutionary view also has
architectural implications.
We aim to survey the existing work on network slicing in the
5G context and identify challenges remaining to be addressed
to make the service-oriented 5G vision a reality. This article
is, to the authors’ knowledge, the first survey on this important
topic. We start by outlining representative 5G architectural
proposals that highlight the crucial role network slicing is
expected to play to meet the widely different requirements of
various use cases. We then present a generic 5G architectural
framework made up of infrastructure, network function, and
service layers as well as the cross-cutting aspect of service
management and orchestration (MANO). With respect to this
framework, we discuss the state of the art on network slicing
in the mobile/5G context. This leads us to identifying some
key outstanding challenges to realize a slice-able, softwarized
5G mobile network architecture.
II. NETWORK SLICING IN THE 5G A RCHITECTURE
This section highlights the need for a flexible 5G architec-
ture to accommodate diverse use cases; outlines representative
architectural proposals that indicate the crucial role network
slicing is expected to play; and presents a generic framework
that broadly represents various proposals and will be used
as the reference for our network slicing literature review in
following sections.
A. Use Cases and Requirements
The 5G network is expected to be the basis for a range
of verticals and use cases. For example, the ITU and 5G-PPP
have identified three broad use case families: enhanced mobile
broadband, massive machine type communications and critical
communications. Within those, it is possible to define several
specific use cases [1] ranging from general broadband access
with global coverage to specialized networks for sensors or
extreme mobility. The stark differences between these use
cases translates to a set of heterogeneous requirements (Fig. 1)
that cannot be satisfied by a one-size-fits-all architecture. With
this in mind, alternative architectural proposals for 5G have
emerged recently to accommodate use cases with diverse
requirements; we outline two such proposals next.

2
Data rate
User Data rate
Spectrum Efficiency
Connection Density
Latency Mobility
Power Efficiency
Traffic Density
Reliability
Machine-to-Machine Critical Communications Mobile Broadband
Fig. 1. Key 5G use cases and their requirements. In this illustration, further
the distance of a requirement from the center, more important it is to the
corresponding use case. It is inspired by a similar illustration from ITU [2].
Diverse use cases need to be mapped to suitably tailored network structures.
It is therefore vital for a 5G architecture to be flexible to realize different
structures as needed.
B. Architecture
NGMN’s architectural vision [1] advocates a flexible
softwarized network approach. This views network slicing as
a necessary means for allowing the coexistence of different
verticals over the same physical infrastructure. Initial proposals
were limited to slicing the CN, but NGMN has argued for an
end-to-end (E2E) scope that encompasses both the RAN and
CN. To realize this and provide the needed context awareness,
both parts need to be flexibly sliced into several overlaid
instances serving different types of users, devices and use
cases. This whole process needs to be orchestrated by an E2E
MANO entity that has a central role in the architecture.
The overall NGMN architecture is split into three layers:
infrastructure resource, business enablement and business ap-
plication. Realizing a service in this proposal follows a top-
down approach via a network slice blueprint that describes the
structure, configuration and work-flows for instantiating and
controlling the network slice instance for the service during
its life cycle. The service/slice instance created based on the
blueprint may be composed of several sub-network instances,
each in turn comprising of set of network functions and
resources to meet the requirements stipulated by the service
in question.
5G-PPP’s architectural vision [3] offers a more elaborate
examination of the roles and relationships between differ-
ent parts of the 5G network. Overall, 5G-PPP shares the
NGMN view that a potential 5G architecture must support
softwarization natively and leverage slicing for supporting
diverse use cases. 5G-PPP’s architectural proposal is divided
into five layers: infrastructure, network function, orchestration,
business function and service layers. Relating this to the
NGMN proposal, while both are built on infrastructure and
network function (business enablement) layers, there are a
couple of differences: the orchestration/MANO is viewed as
a separate layer in the 5G-PPP proposal; and the business
application layer in the NGMN proposal is divided into two
Configuration Life-Cycle
User Plane
Functions
Control Plane
Functions
Radio
Access
Network
Core
Network
(Edge)
Cloud
Service Layer
Enterprise 3rd partyVerticalsOperators
Network Function Layer
Infrastructure Layer
Management & Orchestration
(MANO)
Control Allocation
Mapping
Fig. 2. Generic framework representing various 5G architectural proposals.
We review and appraise the 5G network slicing literature with respect to this
framework.
layers (business function and service) in the 5G-PPP case.
More generally, there seems to be a broad consensus on
the need for native support for softwarization and network
slicing as a means to realize widely different services in
5G. Moreover, various 5G architectural proposals can be
broadly mapped to a generic framework shown in Fig. 2. This
framework is composed of three main layers: the infrastructure
layer, the network function layer and the service (or business)
layer. It also consists of a MANO entity that translates use
cases and service models into network slices by chaining
network functions, mapping them to infrastructure resources,
configuring and monitoring each slice during its life-cycle.
III. STATE OF THE ART ON 5G N ETWORK SLICING
In this section, we discuss the existing work on network
slicing for 5G using the generic framework presented in the
previous section (Fig. 2) as a reference. Table I summarizes
this discussion. Fig. 3 gives an overview of the various
research issues related to network slicing, and indicates where
future research should be focused.
A. Infrastructure Layer
Scope: The infrastructure layer broadly refers to the physical
network infrastructure spanning both the RAN and the CN.
It also includes deployment, control and management of the
infrastructure; the allocation of resources (computing, storage,
network, radio) to slices; and the way that these resources
are revealed to and can be managed by the higher layers.
Existing work: The related work focuses on two main
subjects; the composition of the network infrastructure and
its virtualization.
1) Composition of Network Infrastructure: It has been
advocated that in order to realize slicing we need to move
towards the Infrastructure-as-a-Service (IaaS) paradigm [4],
where different infrastructural elements covering different re-
quirements can be leased to accommodate the needs of the
various slices. This paradigm is well-known in the context of
cloud computing, but it needs to be further adapted for the 5G
context.

3
TABLE I
APPROACHES FOR ADDRESSING DIFFERENT ASPECTS OF NETWORK SLICING AND THEIR (DIS )ADVANTAGES
Advantages Disadvantages
Virtualization of Radio Resources Dedicated Resources
Natrually ensures resource isolation,
ease of realizing virtual base station
stacks and supporting multiple RATs
Inefficient utilization of radio resources
Shared Resources More efficient use of radio resources Requires more sophisticated techniques to ensure
isolation of radio resources
Granularity of Network Functions Coarse-grained Easier deployment and management of
network functions
Less flexible and adaptive to changes in
underlying network conditions
Fine-grained More flexible and easier to conform
to SLAs
Service chaining and interoperability of
functions challenging
Service Description Human-readable format Easier to express service requirements MANO role challenging in mapping requirements to
network components
Set of functions and
network components
Non-intuitive way to express service
requirements Simpler realization of network slices
Considering the core network (CN), there is a broad con-
sensus in favor of using generic hardware infrastructure for
the deployment of virtual network functions [5] [6] [7].
However, due to the differing constraints between various
services deployed over the network, a simple centrally posi-
tioned cloud infrastructure might not be suitable for all the
slices. For example, a sub-millisecond latency requirement
of a tactile service such as remote surgery deployed over a
dedicated network slice cannot be accommodated if the cloud
infrastructure is located far away from the RAN. Therefore,
some network slicing architectures [5] [7] propose a mix
of central and edge cloud computing infrastructures where
resources can be allocated to either of them, depending on
the slice requirements.
On the other hand, the radio access network (RAN) com-
prising of multiple base stations is expected to span diverse
radio access technologies (RATs) including LTE and Wi-
Fi. Moreover, since the slices are expected to be created
dynamically with their service requirements not known a
priori, the RAN infrastructure needs to be flexible enough
to provide support for various RATs on-the-fly, following a
RAN as a Service (RaaS) paradigm. This is why a large
number of architectural proposals [4] [7] [6] for network
slicing expect the deployment of generic software-defined base
stations composed of centralized baseband processing units
and remote radio heads as the logical next step.
2) Infrastructure Virtualization: The ability to virtualize
the underlying infrastructure and provide isolation among
services is essential for network slicing. This not only means
virtualization and full isolation of the underlying resources
(processing, storage, network and radio) among slices but also
the capability to support different types of control operations
over the resources in a virtualized manner based on the service
requirements. This characteristic of providing a virtualized
end-to-end environment, that can be potentially opened up and
fully controlled by third parties, is one of the key features that
separates network slicing from the already existing network
sharing solutions [5] [4].
Considering the virtualization of the CN infrastructure,
research done in the context of cloud computing can be
leveraged. Specifically, technologies like Kernel-based Virtual
Machines (KVM) and Linux Containers (LXC) can provide
isolation guarantees in terms of processing, storage and net-
work resources at the OS or process level. These isolation
guarantees combined with the capabilities offered by platforms
like OpenStack for the pooling of resources can greatly
simplify the on-the-fly creation of virtualized CNs. Due to
the high maturity level of the aforementioned technologies,
concrete prototype implementations of slicing frameworks are
already available, enabling the deployment of virtual core
network functions (virtual MME, virtual SGW, etc.) over cloud
infrastructures (e.g., [7]).
On the other hand, virtualization approaches for the RAN
are at an early stage. Applying VM and container-based
solutions in this domain does not fully address the problem as
they do not deal with the additional dimension of virtualizing
and isolating radio resources (spectrum and radio hardware).
Existing RAN virtualization approaches that account for this
dimension fall into one of two categories: (i) providing a
dedicated chunk of spectrum for each virtual base station
(slice) to deploy a full virtual network stack on top of it
[7]; (ii) dynamically sharing the spectrum between different
virtual base station instances (slices) by employing common
underlying physical and lower MAC layers [8]. The dedicated
spectrum approach is easier to implement, especially with
dedicated radio hardware per slice, since isolation of radio
resources is guaranteed through the static fragmentation of the
spectrum but it can result in inefficient use of radio resources.
The other approach of fine-grained and dynamic spectrum
sharing has the opposite problem of making isolation between
slices challenging.
B. Network Function Layer
Scope: The network function layer encapsulates all the
operations that are related to the configuration and life-cycle
management of the network functions that, after being
optimally placed over the (virtual) infrastructure and chained
together, offer an end to end service that meets certain
constraints and requirements described in the service design
of the network slice.
Existing work: The research interest in this layer mainly
revolves around the technologies that can act as enablers for
the deployment and the management of network functions, as
well as around issues regarding the granularity and the type
of the deployed functions.

4
1) Enabling Technologies: There already seems to be a
consensus among researchers and the industry about the role of
SDN and NFV [7] [9] [6] [5]. NFV is an ideal technology for
the life-cycle management and orchestration of the network
functions, while SDN can inherently act as an enabler of
NFV by allowing the configuration and control of the routing
and forwarding planes of the underlying infrastructure through
standardized protocols (e.g., Openflow).
2) Granularity of Network Functions: One particularly
interesting aspect of this layer that is thoroughly discussed
in various relevant works is the granularity (scope) of the
available virtual network functions [5] [10]. On one end, we
have coarse grained functions, where each one is responsible
for a large portion of the network’s operations (e.g., individual
functions for LTE eNodeBs, MMEs, S-GWs). On the other
end, we have functions with a very fine granularity, where
each of the coarse grained functions mentioned above are
divided further into many sub-functions. For example, in [11]
the LTE EPC is broken into functions responsible for mobility
and forwarding traffic (MME, S-GW, P-GW), which in turn
are further decomposed into sub-functional entities including
signaling load balancers, mobility managers and functions
dedicated to the forwarding of either control or data plane
traffic.
The coarse grained approach offers a more simplified
way of placing and managing the network functions of a
slice. However, this comes at the expense of a slice that is
less flexible and adaptive to the changes of the underlying
network conditions, something that can be critical when the
slice needs to conform to specific service level agreements
(SLAs) [10]. For example, the radio resource scheduler in
a slice might need to be swapped for another one with a
different scheduling policy when a large number of mobile
devices appear concentrated in a specific location in order
to avoid violating the slice’s SLA. If the scheduler is tightly
coupled and packed as a single function with the rest of the
eNodeB, performing the swapping operation can become a
challenge. However, it has also been argued that, despite
its benefits on the adaptation of a slice to the network
conditions, a fine granularity can be problematic for the
interfacing and chaining of the network functions since the
more network functions that exist, the more interfaces need
to be defined for their inter-communication [5]. This is
particularly an issue when virtual network functions are made
available by third-parties through some kind of a network
function store [7], since without common interfaces, their
interoperability is not guaranteed. As a workaround, the use
of a container-based protocol that will wrap the interface of
the contained functions has been proposed [5], however there
is no concrete description as to how to achieve this.
C. Service Layer & MANO
Scope: Perhaps the most important element that distinguishes
network slicing in the context of 5G from other forms of
slicing that have been considered in the past (e.g., cloud
computing) is its end-to-end nature and the requirement to
express a service through a high-level description and to
flexibly map it to the appropriate infrastructural elements and
network functions. This observation regarding the operation
of slicing in the context of 5G naturally leads to two new
high-level concepts: (1) a service layer that is directly linked
to the business model behind the creation of a network slice;
and (2) network slice orchestration for the hypervision of a
slice’s life-cycle.
Existing work: Due to the novelty that this layer introduces
in terms of concepts and ideas, the related research in this
domain naturally focuses on answering fundamental questions
regarding network slicing architectures. More specifically, the
topics considered are related to the way that services should be
described and how they should be mapped to the underlying
network components, and the architecture of network slicing
managers and orchestrators.
1) Service description: Regarding the service layer and the
way that the business model of a service should be described
in high-level terms, there are two different proposals. In
one approach, the service level description (manifest) is
just simply a set of traffic characteristics, SLA requirements
(e.g., for performance related aspects like throughput and
latency) and additional services (e.g., localization service)
[6]. In the second approach, the service description is more
detailed in the sense that it can identify specific functions
or RATs that are bundled together and should be used for
the creation of the slice [7] that provides a specific service
(slice as an application). The main difference lies in the way
that the network slice will be generated. In the first case, the
slice orchestrator will be assigned the more complex task of
identifying the appropriate functions and technologies that
will guarantee the fulfillment of the requirements described
in the slice’s manifest, while in the second case things are
more simplified since the required building blocks of the slice
are already identified in its description. However, the second
approach can be less efficient as it leaves less flexibility to
the slice orchestrator to tune the components of the slice.
2) MANO architecture: The exact form that the network
slicing management and orchestration entity should have is
still unclear with different works presenting different ideas.
Some proposals envision that network slicing will come
through an evolution of the current 3GPP standards and
therefore propose enhancements in terms of interfaces and
functionalities for the existing mobile architecture [12]. Others
envision a more radical clean-slate approach where the slice
management and orchestration will be implemented as an
application over an SDN controller which will oversee both
the wired and the wireless domain [4] [5] [9]. Concrete im-
plementations of MANO reference frameworks such as Open
Source MANO1 (OSM) are already making their appearance,
which enable experimental studies on end-to-end 5G network
slicing.
3) Mapping of services to network components: Another
very important issue is how to map and stitch together the
1https://osm.etsi.org/

5
components that are available to the various layers of the ar-
chitecture in order to compose an end-to-end slice. Two types
of mapping have been considered: (1) the functional/SLA
mapping of the service requirements to network functions and
infrastructure types; and (2) the mapping of network functions
and infrastructure type to vendor implementations [6] [7].
The first type of mapping refers to the way that MANO
chooses appropriate high-level network elements that are re-
quired to create a slice for a given service in order to meet its
functional requirements and SLA. For example, if a slice has
a need to cover devices over a wide area without any capacity
concerns, choosing an LTE deployment with macro cells might
be a good option. For this mapping, it has been proposed that
the available infrastructural elements and network functions
should reveal their capabilities to the MANO in a form of
meta-data, describing the types of services that they can
support [6].
Once the type of functions and infrastructural elements
required for the slice have been identified, there is a need
for a further mapping of these elements to concrete vendor
implementations. Depending on the implementation of a
function by a vendor, different levels of services can be
offered. For example, alternative software implementations of
an LTE eNodeB could provide support for a different number
of users, with different performance guarantees or even with
different capabilities (e.g., flexible modification of the MAC
scheduler). Here too, the high level solution to this problem
seems to be the use of meta-data in the elements provided
by the vendors of the functions and the infrastructure.
Such meta-data could describe both the capabilities of the
vendor specific functions and hardware [6] [5] as well as
their deployment and operational requirements (connectivity,
supported interfaces and infrastructural KPI requirements)
[6] [7], providing the MANO with sufficient information to
perform the best possible configuration for the slice.
IV. CHALLENGES
From the last section, it is apparent that 5G network slicing
has already received fair amount of attention from the research
community and industry. At the same time, there are several
aspects key to end-to-end network slicing that are not well
understood as captured by the illustration in Fig. 3. With this
in mind, we now elaborate on several significant outstanding
challenges that need to be addressed to fully realize the vision
of network slicing based multi-service softwarized 5G mobile
network architecture.
A. RAN Virtualization
As already discussed in Section III-A2, the main challenges
for infrastructure virtualization lie in the RAN. Solutions that
pre-allocate distinct spectrum chunks to virtual base station
instances (slices) are straightforward to realize and provide
radio resource isolation but have the downside of inefficient
use of radio resources. The alternative dynamic and fine-
grained spectrum sharing based RAN virtualization approach
does not have this limitation and therefore is desirable. How-
ever ensuring radio resource isolation is a challenge for this
approach. This can potentially be addressed by adapting SD-
RAN controllers like [13]. As 5G networks are expected to
span multiple RATs (including emerging technologies like 5G
new radio and NB-IoT), it is vital for RAN virtualization
solutions to be able to accommodate multiple RATs. This
presents an additional outstanding challenge as it is unclear
whether multiple RATs can be multiplexed over the same
possibly specialized hardware or each needs its own dedicated
hardware; the answer to this question might depend on the set
of RATs under consideration.
From the RAN virtualization viewpoint, realizing the RaaS
paradigm is another major challenge over and above the ones
outlined above. This is a significant step over the notion of
RAN sharing that involves sharing of radio resources among
tenants (e.g., MVNOs) via a physical mobile network operator;
various solutions for RAN sharing exist (e.g., [8], [13]). RaaS
paradigm requires going beyond radio resource and physical
infrastructure sharing [4] [9] to have the capability to create
virtual RAN instances on-the-fly with tailored set of virtual-
ized control functions (e.g., scheduling, mobility management)
to suit individual slice/service requirements while at the same
time ensuring isolation between different slices (virtual RAN
instances).
B. Service Composition with Fine-Grained Network Functions
Ease of composing a service out of the available network
functions is as discussed in Sec. III-B2 directly dependent on
the granularity of these functions. Coarse-grained functions
are easy to compose as fewer number of interfaces need to be
defined to chain them together but this comes at the cost of
reduced flexibility for the slices to be adaptable and meet their
service requirements. Fine-grained network functions do not
have this limitation and are more desirable. However we lack a
scalable and interoperable means for service composition with
fine-grained functions that could be implemented by different
vendors. The straightforward approach of defining new stan-
dardized interfaces for each new function is not scalable as
the functions increase in number and the granularity becomes
finer.
C. End-to-End Slice Orchestration and Management
A significant challenge for realization of a network slice
is how to go from high-level description of the service to
the concrete slice in terms of infrastructure and network
functions. The problem of describing services has already been
identified in the literature (Sec III-C1) but without satisfactory
resolution. A good approach to address this void is to develop
domain-specific description languages that allow the expres-
sion of service characteristics, KPIs and network element ca-
pabilities and requirements in a comprehensive manner while
retaining a simple and intuitive syntax (e.g., in the philosophy
of [14]). Two important features that such languages should
inherently provide are the flexibility/extensibility to accom-
modate new network elements that may appear in the future
(e.g., new network functions, new RATs) and the applicability

6
Central/Edge Cloud
[5][7]
Issues in RAN
Virtualization
Container-based [7] Virtual Machines [7]
Dedicated Resources
[7]
Shared Resources
[4][6][8][13]
Core Network Core NetworkRadio Access Network
Network Infrastructure
Type Virtualization
Infrastructure Layer
Network Function Layer
Core Network
NFV [5][6][7][9] SDN [5][6][7][9] Coarse Grained [7] Fine Grained [5][11]
Enabling Technologies Network Function
Granularity
Issues in composition
of end-to-end services
MANO
Human-readable
Format [6]
Set of Functions and
Network Components [7]
Architecture
Clean-Slate Approach
(e.g. OSM) [4][5][9]
Evolution of Current
3GPP Standards [12]
Mapping of Services
to Network Functions
and Infrastructure [6][7]
Mapping of Network Functions
and Infrastructure to Vendor
Implementations [5][6][7]
Service Description Mapping of Services to
Network Components
Issues in end-to-end slice
orchestration and management
Mature research with concrete solutions
and/or readily available platforms/tools
Detailed research proposal without
concrete implementation or thorough
evaluation of benefits
Conceptual idea or identified problem
without a detailed solution
Radio Resource
Virtualization RAN as a Service [4][9]
Service Layer
End-to-End management and
orchestration mechanisms [4]
Fig. 3. Maturity level of various aspects of 5G network slicing research.
to be used in multi-vendor environments. A desirable feature
would also be the capability to compose complex rules and
expressions out of simpler ones, introducing abstraction layers
in the expression of service requirements.
As noted in Sec III-C2, concrete MANO frameworks like
OSM have emerged in recent years. While such platforms
are essential to flexibly realize network slices end-to-end and
as needed, there’s a more significant challenge that is only
starting to be addressed. This concerns holistic orchestration
of different slices so that each meet their service/SLA require-
ments while at the same time efficiently utilizing underlying
resources. This calls for a sophisticated end-to-end orches-
tration and management plane. Such a plane should not be
limited to trivial slice generation that does mapping of slices
to network components and statically allocates them resources.
Instead it should be adaptive, ensuring that the performance
and resiliency requirements of the deployed services are met.
To achieve this, it should efficiently and holistically manage
resources by making decisions based on the current state of
slices as well as their predicted state/demands in the near
future [4].
Such issues have been thoroughly investigated in the context
of cloud computing and data centers, where many concrete
solutions have already been proposed (e.g., [15]). While un-
derlying principles from these other contexts can be lever-
aged, mechanisms targeting 5G network slicing should be
suitably adapted and extended considering additional types
of resources. Specifically, not just the resources found in
cloud environments (memory, storage, network), but also radio
resources need to be included, considering their correlation
and how adjusting one resource type could have a direct effect
on the efficiency of some other and therefore on the overall
service quality. The problem of meeting requirements of dif-
ferent services while efficiently managing underlying network
resources in the 5G network slicing context is also somewhat
analogous to the quality of service (QoS) provisioning in the
Internet.
V. SUMMARY
We have presented what we believe to be the first survey
of the state of the art on 5G network slicing. To this end,
we present a common framework for bringing together and
discussing existing work in a holistic and concise manner.
This framework essentially groups existing slicing proposals
according to the architectural layer they target; namely the
infrastructure, network function, and service layers along with

7
the MANO entity. With respect to this framework, we evaluate
the maturity of current proposals and identify remaining gaps.
While several aspects of network slicing at the infrastructure
and network functions layers are quickly maturing, issues like
virtualization in the RAN are unresolved. Also, approaches for
realizing, orchestrating, and managing slices are still in their
infancy with many open research questions.
REFERENCES
[1] NGMN Alliance. 5G White Paper, Feb 2015.
[2] ITU-R. IMT Vision – Framework and overall objectives of the future
development of IMT for 2020 and beyond, Sept 2015.
[3] 5G PPP Architecture Working Group. View on 5G Architecture, Jul
2016.
[4] I. F. Akyildiz, P. Wang, and S. Lin. SoftAir: A software defined
networking architecture for 5G wireless systems. Computer Networks,
85(C), 2015.
[5] P. Rost et al. Mobile network architecture evolution toward 5G. IEEE
Communications, 54(5), 2016.
[6] X. Zhou, R. Li, T. Chen, and H. Zhang. Network slicing as a service:
enabling enterprises’ own software-defined cellular networks. IEEE
Communications, 54(7), 2016.
[7] N. Nikaein et al. Network store: Exploring slicing in future 5G networks.
In Proc. 10th ACM International Workshop on Mobility in the Evolving
Internet Architecture (MobiArch’15), Sep 2015.
[8] X. Costa-Pérez, J. Swetina, T. Guo, R. Mahindra, and S. Rangarajan.
Radio access network virtualization for future mobile carrier networks.
IEEE Communications, 51(7), 2013.
[9] A. Banchs, M. Breitbach, X. Costa, U. Doetsch, S. Redana, C. Sar-
tori, and H. Schotten. A novel radio multiservice adaptive network
architecture for 5G networks. In Proc. 81st IEEE Vehicular Technology
Conference (VTC Spring’15), May 2015.
[10] M. Sama, X. An, Q. Wei, and S. Beker. Reshaping the mobile core
network via function decomposition and network slicing for the 5G era.
In Proc. IEEE WCNC, Apr 2016.
[11] F. Z. Yousaf, P. Loureiro, F. Zdarsky, T. Taleb, and M. Liebsch. Cost
analysis of initial deployment strategies for virtualized mobile core
network functions. IEEE Communications, 53(12), 2015.
[12] K. Samdanis, X. Costa-Perez, and V. Sciancalepore. From network
sharing to multi-tenancy: The 5G network slice broker. IEEE Com-
munications, 54(7), 2016.
[13] X. Foukas, N. Nikaein, M. M. Kassem, M. K. Marina, and K. Konto-
vasilis. FlexRAN: A flexible and programmable platform for software-
defined radio access networks. In Proc. ACM CoNEXT, Dec 2016.
[14] Q. Zhou, C. Wang, S. McLaughlin, and X. Zhou. Network virtualization
and resource description in software-defined wireless networks. IEEE
Communications, 53(11), 2015.
[15] D. Shue, M. J. Freedman, and A. Shaikh. Performance isolation and
fairness for multi-tenant cloud storage. In Proc. 10th USENIX Sympo-
sium on Operating Systems Design and Implementation (OSDI’12), Oct
2012.