Wireless Pers Commun (2014) 78:1629–1644

DOI 10.1007/s11277-014-1909-6

SDN Based Mobile Networks: Concepts and Benefits

Slavica Tomovic · Milica Pejanovic-Djurisic ·

Igor Radusinovic

Published online: 11 July 2014

© Springer Science+Business Media New York 2014

Abstract Mobile Internet traffic is expected to grow faster than the fixed Internet traffic in
a near future. Since, today’s broadband networks are approaching theoretical limits in terms
of spectral efficiency per link, mobile operators are compelled to seek for new solutions that
will be able to accommodate the expected traffic growth and improve their position in the
competitive market. Addressing those challenges with current inflexible, not scalable and
complex architecture is very hard, if possible at all. Software defined networking (SDN)
is a new networking architecture paradigm that holds great promise to overcome many of
mentioned limitations and provides required improvements in performance by decoupling
control functions from underlying physical infrastructure. In this paper, we explain key rea-
sons for transition to SDN based mobile networks and briefly describe several proposals of
design scenario. Special emphasis is placed on SDN’s contribution to more efficient inter-cell
interference management, traffic control and network virtualization.

Keywords Mobile networks · SDN · Traffic control · Network virtualization

1 Introduction

The last few years have witnessed a tremendous growth of mobile data traffic, in large part
due to increasing popularity of smart phones, tablet computers and Machine-to-Machine
(M2M) devices [1]. This trend is expected to continue in the future. According to Cisco’s
VNI (Visual Networking Index) 2013 forecast, mobile data traffic will increase 13 times
between 2012 and 2017 [2]. Such pace of growth, inherent need for simultaneous support of

S. Tomovic (B) · M. Pejanovic-Djurisic · I. Radusinovic

Faculty of Electrical Engineering, University of Montenegro, Podgorica, Montenegro
e-mail: [email protected]
M. Pejanovic-Djurisic
e-mail: [email protected]
I. Radusinovic
e-mail: [email protected]

123
1630 S. Tomovic et al.

multiple wireless technologies and constant introduction of new services impose significant
challenges for today’s cellular network architecture. In order to increase capacity through
frequency reuse, operators have to deploy dense cell environment in the radio access networks
(RANs). Reduced distance between user and the access infrastructure entails better link
quality and smaller number of users that each base station needs to support. However, due
to lack of available spectrum, base station operating on the same frequency channel become
more closer to each other, thereby causing higher level of inter-cell interference. Smaller
coverage areas also mean more frequent handovers, cell association and resource allocation
which are very hard to manage in networks designed on traditional way, where each base
station mostly makes independent control plane decisions on the radio layer [3].
Another trend is heterogeneous network environment with diverse wireless technologies
that are used in parallel (3G, 4G, Wi-Fi, Bluetooth). This necessitates numerous operational
staff and deployment of very complex policies in order to guarantee right access to right
services and control handovers between different access types. Also, protocols in today’s
wireless networks are tightly coupled with the hardware in the form of ASICs (Application-
Specific Integrated Circuits) designed for each of them, so any important protocol update
requires upgrade of base stations or even their replacement. In dense cell scenario this is
prohibitively expensive. “Closed” nature of modern networking equipment prevents mobile
operators to test and deploy new features and consequently makes them unable to respond to
quickly changing user needs. This is especially important when considering more and more
powerful competitors, such as over-the-top (OTT) players, that today’s mobile operators must
cope with by offering innovative and attractive services to customers. Taking into account
all these issues, it can be concluded that the carrier network paradigm has reached its limits
and is not able to satisfy stringent market requirements.
Complex, inflexible and functionally limited architecture is not typical just for mobile cel-
lular networks. It applies more generally to all networking technologies. There, three major
culprits are mostly seen as: (i) coupling of the data plane and the control plane at the node
level what makes network equipment hard to configure optimally and troubleshoot, (ii) lack
of useful abstractions and modularization of the control plane which can facilitate network
management and development of new services, and finally, the most important which is
highly correlated with the previous two: (iii) lack of programmability of the network as a
whole [17]. On a such ground, the idea about SDN has emerged, aiming to provide fully
programmability instead of limited configurability we have today. The SDN concept is based
on split architecture, where forwarding functions are decoupled from control functions. Net-
work intelligence is shifted to logically centralized SDN controller, which maintains global
view of the network, interacts with simple switching devices and provides a programming
interface for user-written network management applications. Consequently, applications and
policy engines have abstracted view on the topology, which for them appears as a single logi-
cal entity. Management of such networks is not any more limited to configuration options that
equipment vendor provides. Operator is free to program devices and alter network behaviour
in real-time by automated SDN programs.
In this paper, implementation scenarios of SDN in mobile cellular networks are examined.
Special attention is payed on SDN’s contribution to more efficient inter-cell interference
management, traffic control and network virtualization. Thus, in Sect. 2 brief background
on today’s mobile broadband networks is provided. After short presentation of SDN and
OpenFlow [4] in Sect. 3, an overview of several architectural proposals for SDN-based
mobile networks is given in Sect. 4. Then, a few use cases that would especially benefit
from SDN paradigm are described in Sect. 5. Section 6 presents market status and related
challenges. The paper conclusions are given in Sect. 7.

123
SDN Based Mobile Networks 1631

Fig. 1 EPS key elements and interfaces

2 Distributed Architecture of Traditional Broadband Mobile Networks

Motivated by the increasing throughput and Quality of Service (QoS) requirements, 3GPP
proposed all-IP (Internet Protocol) network core Evolved Packet Core (EPC) [5]. EPC, along
with the Evolved Universal Mobile Telecommunication System (UMTS) and Terrestrial
Radio Access Network (E-UTRAN), forms Evolved Packet System (EPS) architecture that
serves as a basis for the fourth generation mobile cellular networks.
Figure 1 illustrates the key entities and interfaces that participate in typical EPS opera-
tion. Traffic from UE (User Equipment) is directed through a set of GTP (GPRS Tunnelling
protocol)/PMIP (Proxy Mobile IP) tunnels from the attached eNB (E-UTRAN NodeB) over
the serving gateway (SGW) and packet data network gateway (PGW), respectively, to the
corresponding packet data network (PDN). SGW routes incoming and outgoing IP packets
from/to UEs inside its working domain. SGW also must track users location and maintain
large databases since users retain their IP addresses when they move. The PGW acts as inter-
face between EPC and external IP networks. It is responsible for packet filtering, enforcing
of various policies and charging support. PGW and SGW also participate in several control
plane protocols through information exchange with MME (Mobility Management Entity)
and PCRF (Policy and Charging Rules Function) entities. MME is a key control plane ele-
ment that performs signalling related to mobility, access control and QoS negotiation. MME
allocates necessary resources on a per-session level and manages mobility state in order to
support tracking, roaming and handovers. PCRF identifies flow to which service data belongs
on the basis of the user subscription profile stored in HSS (Home Subscriber Server) database,
and accordingly manages charging and sends instructions to PGW for enforcing appropriate
policy over that traffic flow.
Although this architecture offers significantly improved network performances, huge pop-
ularity of mobile communications raised new challenges. One of the most important is related
to scalability, because EPS architecture implies that all traffic goes through PGW, even if it is
between users within the same cellular network. Centralization of monitoring, access control
and QoS functionalities at PGWs also entails very high price of these devices (e.g. more than
6 millions dollars for Cisco packet gateway) [6]. Vendor dependence is another problem.

123
1632 S. Tomovic et al.

Fig. 2 SDN architecture

New functionalities can only be added by developers employed by vendors, so introduction

of new services often requires replacement of the network equipment.

3 Software Defined Networks and OpenFlow

In traditional telecommunications networks control and data plane are combined in each
device. The control plane carries signalling traffic, performs route calculation, system con-
figuration and management. It contains all the logic that controls the behaviour of the network.
On the other hand, the data plane is focused only on packet transport towards their next des-
tination. The network itself basically can be seen as a distributed entity that connects diverse
independent and autonomous devices. Such architecture does not provide abstracted view of
the network for applications and network services, which resulted in its high complexity, as
one of the key deficiencies. Management is done at a very low level, and once forwarding
policy has been defined, the only way to make an adjustment to the policy is via manual con-
figuration of the devices. Network administrators have to react to every network event and
due to this they often become performance bottleneck in situations when amount of traffic
rapidly increases. Such inflexible architecture, which is slow and costly to manage, limits
ability to introduce new services in the network or to adapt the network to requirements of
new services.
SDN is a new approach in networking technology, designed to improve the complex and
static nature of traditional distributed network architectures. The SDN concept is based on
split architecture where control plane is separated from underlying physical infrastructure.
Independent and centralized unit to control the network is provided. Since all control functions
are moved to a programmable controller, administrators can easily alter network behaviour
in real-time and adapt it to quickly-changing network conditions by running appropriate con-
troller applications. Three-layer model that SDN uses is illustrated in Fig. 2. Communication

123
SDN Based Mobile Networks 1633

between different layers is realized through two interfaces: (i) the northbound interface which
connects applications to the controller (REST, JSON, XML, etc.) and (ii) the southbound
interface which handles all communications between controller and data-forwarding plane.
So far, OpenFlow protocol is the most commonly used southbound interface. It provides data
plane abstraction by introducing concept of flow table. Flow table resides on each network
device and contains a set of rules for traffic flow processing, installed dynamically or stati-
cally by the controller. Flow can be defined with different level of granularity by choosing
any subset of l2 L2-L4 fields supported by specification, alongside with ingress port number.
As OpenFlow provides possibility for configuring network on a per-flow level, network can
be flexibly adjusted to applications and user requirements. Current networks based on IP pro-
tocol do not offer this level of control, because they imply that all traffic flows between two
ends of systems traverse the same path. Beside this, SDN significantly contributes to lower
capital expenditures since network equipment no longer need to understand and process huge
number of protocol standards but only to act according to controller’s instructions. When a
packet enters an OpenFlow enabled network device, network device tries to classify it in one
of the existing flows. If no match is found, packet will be sent to the controller. Applications
that run on top of the controller decide how corresponding flow should be handled. After
decision making, new entry is filled in device’s flow table. Beside flow definition, this entry
contains a list of actions and statistical counters. These actions allow to manipulate with
headers, to forward packets to output ports or specific queues on them, to drop packets or
encapsulate and send them to the controller.

4 Architecture Proposals for SDN Based Mobile Cellular Networks

Although SDN architecture was designed with wired networks in mind, implementation of
SDN technology in mobile domain is gaining more and more attention from research com-
munities. Development of “programmable mobile networks”, that allow greater flexibility in
management and configuration, is seen as a step in right direction for overcoming shortcom-
ings of traditional mobile networks. Therefore, in the remainder of this section we provide
overview of different projects dealing with mobile cellular networks that incorporate SDN
concept.
One of such projects is SoftCell [7], which provides framework for incorporating SDN
principles in 4G LTE (Long Term Evolution) core networks. The key idea is to create mobile
network consisted of simple programmable core switches, with most of the functionalities
moved to the access switches (close to the base station). This partition on access and “other”
core switches is done because implementation of pure SDN architecture (that implies com-
pletely absence of control functions on data plane) in mobile cellular networks introduces
some big scalability challenges. For example, due to users mobility, data plane will gener-
ate a large number of state updates, which would cause controller congestion. In order to
achieve satisfying average response latency it is necessary to reduce amount of control traf-
fic that controller and network devices exchange. Because of that, in SoftCell fine-grained
packet-classification on traffic from UEs is done on access switches only. Since an access
switch has to handle only traffic from the base station associated to it, this solution is much
more scalable than traditional LTE design where a few PGWs handle traffic from all base
stations. Also, they are envisioned to perform some simpler control-plane functions in order
to minimize interaction with the centralized controller. In general, principle of operation is
as follows. When packet arrives at access switch, if it does not match any of entries in flow
table it will be sent to the controller. Controller has global view of the network, and on the

123
1634 S. Tomovic et al.

Fig. 3 SoftCell architecture [7]

basis of subscriber’s profile and application requirements enforces various service policies
by installing switch-level rules that direct traffic through differently dedicated appliances
(middleboxes), as illustrated in Fig. 3. Forwarding rules contain three match fields: policy
tag, base station (hierarchical IP address) and UE identifiers, whereby not all of them have
to be specified. This aggregation on three dimensions significantly reduces the size of switch
tables. To avoid reclassifying the return traffic at gateway switches, SoftCell embeds loca-
tion and policy information in the packet header, precisely in place of source IP address and
source port number so traffic that comes from Internet will carry these fields for sure.
Opposite to SoftCell that focuses at redesigning the core network, SoftRAN [3] uses SDN
principles to redesign the LTE radio access network. Currently, radio access networks use
distributed protocols to manage interference and perform handovers. While this solution is
acceptable in sparse environment, in dense networks it leads to poor performances since
communication among large number of base stations, that is necessary for efficient handling
of interference and mobility, introduces serious latency. In SoftRAN whole LTE network
is controlled in a centralized way: all base stations deployed in local geographical area
are treated as a virtual big-base station, which is made up of radio elements (the individual
physical base stations) and managed by SDN controller. Radio resources that network possess
on its disposal are abstracted out as three dimensional grid of space, time and frequency slots.
SDN controller periodically receives information about local network state from all the radio
elements and based on this it decides where to assign which time-frequency resource block
and defines transmitting power for each of them. As shown in Fig. 4, while making a decision,
the controller takes into account interference level on communication channels, flow statistics,
subscriber’s profile and preferences of network operator. SoftRAN copes with the inherent
delay between the centralized controller and individual radio elements by shifting to radio

123
SDN Based Mobile Networks 1635

Fig. 4 SoftRAN architecture [3]

elements control tasks for which only knowledge of local network parameters is necessary. On
the other side, all control decisions that affect global network state (e.g. handovers, transmit
power settings) are made by centralized controller. Such organization of control plane enables
the network to adjust to rapidly varying channel conditions.
MobileFlow [8] arhitecture is designed in way to facilitate migration from traditional to
software defined networks. As illustrated in Fig. 5, it consists of MobileFlow forwarding
engines (MFFE), which are interconnected by IP/Ethernet transport network, and Mobile-
Flow controller (MFC). In compliance with basic SDN principles, mobile network control
is centralized at MFC and separated from data plane elements (MFFEs). MFFEs are fully
software driven, i.e. its behaviour defines MFC through appropriate API (Application Pro-
gramming Interface). For example, applications that run on top of the MFC can implement
the control functionality of each of the EPS elements, which means that operator can make
high-performance broadband network without complex devices such as E-NodeBs, PGWs
and SGWs. In a similar way, any other existing or new mobile technology can be deployed.
IP/Ethernet transport network is software-defined too, and it uses OpenFlow protocol for com-
munication between controller and switches. However, capabilities that OpenFlow provides
are not sufficient for mobile network data plane. MFFEs support network layer tunnelling
and flexible charging. This makes them more complex than OpenFlow switches, but also
much simpler than traditional EPS elements since majority of control plane functionality is
shifted to MFC. Thanks to the tunnel processing ability, MFFEs, with the support of an MFC,
can interact with legacy EPC elements.
The project OpenRadio [9] proposes design for a programmable wireless data plane that
enables implementation of various wireless protocols on the same hardware platform (e.g.
DSPs, FPGAs, multi-core desktop machines). The long-term goal of this project is to realize
software defined cellular infrastructure based on programmable base stations, backhaul net-
work switches/routers and gateways. The key idea is to decouple the protocol functionality
into two parts: (i) processing plane comprised of the set of algorithms that process data using
underlying hardware, and (ii) decision plane which contains the logic that dictates which

123
1636 S. Tomovic et al.

Fig. 5 MobileFlow architecture [8]

Table 1 Processing plane blocks common for different technologies [9]

Processing plane block WiFi LTE 3G DVB-T

√ √ √ √
FIR/IIR
√ √ √ √
Correlation
√
Spreading
√ √ √
FFT
√ √ √ √
Channel estimation
√ √ √ √
QAM mapping
√ √ √ √
Interleaving
√ √ √ √
Convolutional coding
√ √
Turbo coding
√ √ √ √
Randomization
√ √ √
CRC

subset of algorithms will be used for a particular packet processing and in which order. This
is especially convenient since different wireless protocols have in common many heavy com-
putational processing plane blocks at the PHY and MAC layers (Table 1). OpenRadio can
be considered as combination of SDR (Software Defined Radio [10]) and SDN. It decou-
ples service definition from underlying hardware, what is in compliance with SDN goals,
but focus is on wireless data plane and providing flexibility at PHY and MAC levels. The
processing plane corresponds to signal processing chain comprised of modules that perform
FFT (Fast Fourier Transform), 64-QAM (Quadrature Amplitude Modulation) mapping, CRC

123
SDN Based Mobile Networks 1637

Table 2 Comparison of different SDN related projects dealing with mobile cellular networks

Project name Requirements The key benefits

SoftCell Modification of the network core Flexible routing

Real-time monitoring
Streamlined operational environment
Reduced expenditures
Rapid innovations
SoftRAN Modification of the access network Better radio resource management
Better mobility support
Efficient interference coordination
Traffic offloading
Reduced expenditures
MobileFlow Modification of the core and radio access Capability of interaction with legacy
networks EPC network elements
Support for GTP tunneling at programmable Software definition of different
data plane mobile networks (3G, 4G and others)
Support for fine-grained policy
enforcement
OpenRadio Replacement of the wireless data plane Implementation of various wireless
protocols on commodity multi-core
platforms
Support for application-specific
wireless service
Modular and declarative interface to
program wireless protocols

(Cyclic Redundancy Check), convolutional encoding, Viterbi decoding, etc. Service defini-
tion is separated from the hardware on which these modules are implemented. The decision
plane defines rules to match subset of traffic flows on the basis of packet header fields, and
specifies actions to process them in form of processing plane blocks that will be used. With
such programmatic control, user and applications can archive the connectivity they need from
whatever wireless access points are around.
In order to have better insight into benefits and shortcomings of the above described
projects, their comparison is given in Table 2.

5 Benefits of SDN

In this section a few specific use-cases are described in order to illustrate key benefits provided
by SDN implementation in mobile networks. Thus, it is chosen to elaborate the following
major improvements: more efficient inter-cell interference management, easier traffic control
and efficient network virtualization.

123
1638 S. Tomovic et al.

5.1 Inter-cell Interference Management

One of the basic strategies that mobile operators use today to cope with capacity problems in
their networks is cell-size reduction. Implementation of small cells enables better frequency
reuse and hence better spectral efficiency. However, this improvement clearly comes at the
cost of higher inter-cell interference. The cellular industry has recognized the difficulty in
managing interference in heterogeneous networks, and developed numerous techniques to
address this issue. Techniques used in the existing LTE networks are: Inter-cell interference
coordination (ICIC), Enhanced inter-cell interference coordination (e-ICIC) and Coordi-
nated multi-point transmission/reception (COMP). These techniques are based on complex,
distributed algorithms with negative influence on the performance of the system in many
aspects. For example, COMP reduces interference at the cell edge by coordinated scheduling
of several neighbouring cells or by joint transmission so that the quality of service for cell-
edge users can be improved. This approach introduces significant processing overhead and
imposes higher power and network resource requirements on the radio access network [11].
SDN can handle interference problem in a better way. Since logically centralized controller
provides insight into the global network state, including information about the current power
and subcarrier allocation profile of all base stations, radio resource allocation decisions can
be made far more optimally. In addition, separation of control and data planes enables new
radio resource management techniques to be easily deployed without modification of the
networking hardware.

5.2 Mobile Traffic Management

SDNs flow-based forwarding model and centralized control provide greater flexibility in
traffic management than traditional mobile networks. Some of the beneficiary use-cases are:
traffic offloading, service discrimination and traffic optimization.
Offloading is also known as Wi-Fi roaming as it implies moving traffic from a mobile net-
work to Wi-Fi network. Rules that trigger offloading action can be defined by either end-user
or an operator. End users usually do this to control data service cost or because of higher
bandwidth availability, while operators try to reduce congestion of the base stations. However,
in traditional mobile networks, there is no protocol that controls forwarding across differ-
ent cellular technologies. Consequently, handovers between different technologies involve
complex procedures that introduce long delays and high packet loss rates. Traffic can not
be repositioned dynamically, so offloading in today’s networks does not contribute to better
resource utilization as much as it should. In SDN networks situation is different. SDN pro-
vides control protocol (e.g. OpenFlow) that works across different technologies, so seamless
handover prerequisite is achieved. Offloading can be done dynamically and on various trigger
criteria, such as: number of users per base station, available bandwidth, subscriber profile,
application type and flow duration. Also, each flow-table entry in SDN switches contains
byte and packet counters which allow operator to use as a triggering criteria individual flow
rate or an aggregate flow rate per application, cell, user, etc.
Another important contribution of SDN for traffic management is capability to enforce
fine-grained policies in distributed fashion. In today’s LTE networks, base stations enforce
only simple policies such as maximum rate regulation. For all other policy tasks is responsible
PGW, which classifies traffic from all base stations and either drop packets, if they violate
firewall policy, or map them into QoS classes by setting up appropriate Diff-Serve code in the
IP header. Due to unbalanced distribution of duties among network elements, system suffers
from poor scalability and underutilized bandwidth inside the network. On the other side, in

123
SDN Based Mobile Networks 1639

SDN, policy enforcement tasks can be distributed over multiple switches. Even more, while
the number of QoS classes that DiffServ mechanism uses is limited, in SDN it is possible to
differentiate network traffic based on the individual flow requirements.
With flexible, dynamic traffic steering, traffic optimization and system protection can
be performed more efficiently. Today, mobile carriers use middleboxes (e.g. firewalls,
transcoders, echo-cancellers) to improve video quality, security, remove echo from voice
communication and so on. However, since they do not have fine-grained control over rout-
ing, traffic is often directed through unnecessary middleboxes. SDNs capability to control
traffic on the per-flow basis enables more optimal utilization of this expensive equipment.
According to its requirements, traffic flow can be steered through needed middleboxes and
bypass all the others, which results in CAPEX (CAPital Expenditures) savings owing to
the avoidance of over-dimensioning. Some of middleboxes can be even replaced by SDN
switches (e.g. load balancer, firewall, network address translator), although due to lack of sup-
port for more complex tasks, such as those performing deep-packet modifications, dedicated
physical appliances will still be needed.

5.3 Network Virtualization

The research community all over the word has widely recognized network virtualization
as one of the fundamental features of the next generation networks. Its importance lies in
capability to create multiple logical networks on top of the same physical infrastructure.
Today, users often experience poor quality of service while using wireless networks on the
go. Throughput and latency are unpredictable, faults are frequent and even connectivity is
sometimes questionable. Ironically, this all happen while a user is surrounded by multiple
WiFi and cellular base stations and its device support variety of wireless technologies. Each
mobile operator uses separate physical infrastructure and by contract restricts its customers
in terms of access to other networks. On the other side, most of private WiFi networks
require authentication and cannot be accessed by anybody. Common virtualization solutions
(e.g. MPLS/BGP VPN technologies) can efficiently isolate traffic form different enterprise
customers but do not provide a straightforward definition of a virtual network. The full
realization of virtual networking concept requires high-level hardware abstraction and that
is exactly what SDN provides [12].
With SDN virtualization solutions, such as FlowVisor [13], multiple tenants can share a
single switch data plane in a controlled and isolated manner (Fig. 6). FlowVisor is specialized
SDN controller that completely virtualizes network by using OpenFlow protocol as hardware
abstraction tool. Network virtualization is done by partitioning link bandwidth, topology,
traffic, device CPU and flow tables. Above the hardware abstraction layer, each of the isolated
network slices can run its own protocols, i.e. can be optimized for the applications running
on it and customized for the operator to which they belong. All traffic management decisions
are performed by programmable entity—OpenFlow controller, whereby FlowVisor serves
multiple controllers—one per each slice. Each controller sees and controls its slice only.
FlowVisor defines a slice as a set of flows running on a topology of switches and ensures
isolation by transparently modifying messages which OpenFlow controllers and switches
exchange. Extending FlowVisor to support base stations virtualization will lead to a more
efficient utilization of the scarce wireless resources. Additionally, since multiple operators
can use the same radio access network, smaller amount of radio equipment will be required
which entails smaller energy consumption of wireless networks. Virtualization of base station
can be accomplished in several ways. For example, slicing can be done in time domain and
each of virtual base stations can use all subcarriers and maximum allowed transmission power

123
1640 S. Tomovic et al.

Fig. 6 Network virtualization with FlowVisor

during time slots that are assigned to them, or they can get different subsets of subcarriers
and perform transmission during all time slots with a fraction of the maximum transmission
power.

6 Market Positioning and Remaining Challenges

As new technologies such as content distribution and cloud computing enter the mobile
networks domain, SDN finds more and more space for implementation. A large number of
equipment vendors have publicly discussed their SDN strategy, including Cisco, Juniper,
Huawei, Infinera, Alcatel-Lucent and Ciena. According to Informa recent survey, a vast
majority of operators believe that SDN is essential for the future of networks [14]. However,
there are many different opinions how and where in the network SDN should be initially
deployed.
One of the biggest immediate benefits that SDN brings to mobile operators is network
virtualization, which implies replacement of dedicated, proprietary hardware with commodi-
tized, completely open IT equipment, and implementation of all control functionality in soft-
ware. Not all network areas are equally suitable for virtualization, although most are highly
desirable. As shown in Table 3, virtualization of the network edge is most beneficial since it
greatly simplifies process of the access technology upgrading. In that way, mobile operators
can cope with rising user demands in the future without hardware replacement, but required
initial investments are prohibitively large. This also comes at wrong time, since the LTE
networks have just been deployed and until operators experience some significant revenues
from them, they will not be keen on investing billions of dollars again. Cloud RAN (C-RAN
[15]) from China Mobile is first attempt of the radio layer virtualization where all baseband
units are integrated as virtual machines on the same server, supporting various radio access
technologies. Although this solution promises improvements in both spectral and energy
efficiency, high expenditures it requires can be justified only when small cells become more
widespread. Initial implementations of mobile network virtualization are more expected in
the EPC which is far more centralized compared to RAN and flat (IP) in its nature. In this
scenario most of the network core functionalities (e.g. MME, SGW and PGW) can be moved

123
SDN Based Mobile Networks 1641

Table 3 Virtualization benefits for mobile networks [14]

Network layer Cost Virtualization benefits

Radio High Less reliance on proprietary hardware

More flexibility to launch new air interfaces
Fluid and scalable baseband processing (Cloud RAN)
Core Medium Adaptive traffic management
Ability to implement components in data center
Less reliance on overprovisioning
Gradual introduction of new elements and services
BSS/OSS Low Service exposure
Billing system flexibility and scalability (e.g. online charging)
Easier introduction of mobile virtual network operators (MVNOs)

into the cloud, which would make the mobile networks more scalable, open to innovation,
and thereby more efficient. For this reason, most mobile operators see SDN primarily as
technology that will accelerate adoption of cloud computing paradigm in their networks. At
the end, the BSS (Business Support System)/OSS (Operations Support System) could be
easy for virtualization, but SDN does not solve any of critical challenges and may not lead
to considerable cost savings.
Implementation of SDN in wireless mobile networks is still at an early stage and a number
of challenges remains to be addressed. Since the time when SDN concept first appeared, there
has been a lot of discussion about controller scalability and security issues. When it comes
to controller scalability, primary concern is the latency introduced by information exchange
between a large number of network devices and a single controller. Deployment of multiple
controllers is seen as a solution, but it does not eliminate the challenge of controller-to-
controller communication which requires global network view. Security problems are also
highly related to centralization of control functions, since any intrusion at SDN controllers
could impact the whole network. In addition, implementation of SDN in wireless domain
poses a lot of radio specific problems. In contrast to the wired networks where FlowVisor
can in rather simple way to isolate multiple virtual networks, most of the current wireless
SDN implementations only work well when slicing uses different channels. Due to limited
number of independent communications channels (e.g. the three non-overlapping in Wi-Fi
2.4 GHz band), FlowVisor has to plan geographical channel reuse in a large networks. Time
division multiplexing can be used as alternative, but it requires a fine coordination of close
access points and very accurate synchronization so that time slots do not overlap. Beside this,
two more challenges are identified in [16]. The first is estimation of channels loads and links
characteristics in terms of delay, loss rate and stability. It is essential for SDN controller to
make efficient resource allocation decisions. The channel load is very difficult to estimate,
as wireless channel status is affected by various factors, ranging from multipath fading,
shadowing and interference from neighbouring access points to small environmental events.
Exhaustive reports about all these variations would unnecessary congest control, because
not all of them are relevant. So, it is important to determine a satisfactory level of status
report thoroughness that will not slow down too much detection of important changes in the
environment. The other challenge is related to a new type of handover that SDN introduces—
handover between service providers. Here, implementation of mobile IP is seen as a solution
that could enable user to seamlessly attach to any access point around.

123
1642 S. Tomovic et al.

There is no doubt that solving the above mentioned issues, even partially, would result in
better utilization of spare wireless spectrum and improved quality of service.

7 Conclusion

Today’s mobile networks are based on complex, pricey and proprietary equipment that do
not provide satisfying flexibility of the network control functions from the point of accom-
modating the expected traffic growth and improving operator’s position in the competitive
market. This paper describes potential benefits and contributions that SDN paradigm can
bring to overcome many of such limitations and to provide required improvements in the net-
work architecture by decoupling control functions from underlying physical infrastructure. In
order to summarize majority of the work done in this relatively new research area, we briefly
described several architectural proposals for SDN based mobile cellular networks presented
in literature. Further on, we have shown that key possible benefits achieved by introducing
SDN concept in mobile networks are: more efficient inter-cell interference management,
easier traffic control and full realization of network virtualization. Namely, SDN can handle
interference problem in a better way with optimal radio resource allocation decisions, since
SDN network controller provides insight into the global network state, including informa-
tion about current power and subcarrier allocation profile of all base stations. In addition, by
separation of control and data planes, new RRM techniques will be easily deployed with-
out modifications of the networking hardware. Easier traffic control with SDN tool will be
provided through: traffic offloading, service discrimination and traffic optimization. Finally,
it has been shown that implementation of SDN virtualization solutions could provide full
base station virtualization that will lead to a more efficient utilization of the scarce wire-
less resources. However, SDN is still in its infancy and on the way towards SDN based
mobile networks a number of issues remain to be addressed. Some of them are related to
controller scalability, security and virtualization in wireless domain. Important challenge is
development of SDN applications that will take advantage of centralized control for: mobility
management, interference coordination and seamless handover between service providers.
Since SDN represents revolutionary approach to the network design, despite all benefits that
it could provide, the success of this idea largely depends on the interests and activity of the
research community.

Acknowledgments The work presented in this paper has been performed within the activities of the EU
FP7 For e-Mont project

References

1. Shafiq, M. Z., Ji, L., Liu, A. X., Pang, J., & Wang, J. (2012). A first look at cellular machine-to-machine
traffic: Large scale measurement and charecterization. ACM Sigmetrics, 40(1), 65–76.
2. Cisco. (2013). Cisco visual networking index forecast projects 13-fold growth in global mobile internet
data traffic from 2012 to 2017. http://newsroom.cisco.com/release/1135354. Accessed 22 February 2013.
3. Gudipati, A., Perry, D., Li, L. E., & Katti, S. (2013). SoftRAN: Software dened radio access network. In
ACM SIGCOMM HotSDN workshop (pp. 25–30).
4. McKeown, N., Anderson, T., Balakrishnan, H., Parulkar, G., Peterson, L., Rexford, J., et al. (2008). Open-
Flow: Enabling innovations in campus networks. ACM SIGCOMM Computer Communication Review
(CCR), 38(2), 69–74.
5. 3GPP. (2008). Description document of LTE Release 8. http://www.3gpp.org/specifications/releases/72-
release-8. Accessed 22 February 2013.

123
SDN Based Mobile Networks 1643

6. Li, L.E., Mao, Z. M., & Rexford, J. (2012). Toward software-defined cellular networks. In European
workshop on software defined networking (EWSDN) (pp. 7–12).
7. Jin, X., Li, L. E., Vanbever, L., & Rexford, J. (2013). Softcell: Scalable and exible cellular core net-
work architecture. In ACM conference on emerging networking experiments and technologies (CoNEXT)
(pp. 163–174).
8. Pentikousis, K., Wang, Y., & Hu, W. (2013). MobileFlow: Towards software-defined mobile networks.
IEEE Communications magazine, 51(7), 44–53.
9. Bansal, M., Mehlman, J., Katti, S., & Levis, P. (2012). OpenRadio: A programmable wireless dataplane.
In Hot topics in software defined networks (HTSDN) (pp. 109–114).
10. Ulversoy, T. (2010). Software dened radio: Challenges and opportunities. IEEE Communications Surveys
& Tutorials, 12(4), 531550.
11. Open Networking Foundation. (2013). OpenFlow-enabled mobile and wireless networks. ONF solu-
tion brief. https://www.opennetworking.org/images/stories/downloads/sdn-resources/solution-briefs/
sb-wireless-mobile. Accessed on 22 February 2013.
12. Yap, K.-K., Sherwood, R., Kobayashi, M., Huang, T.-Y., Chan, M., N. Handigol, N., et al. (2010). Blueprint
for introducing innovation into wireless mobile networks. In ACM SIGCOMM workshop on virtualized
infrastructure systems and architectures (VISA10) (pp. 25–32).
13. Sherwood, R., Gibb, G., Yap, K. K., Appenzeller, G., Casado, M., McKeown, N., et al. (2010). Can the
production network be a testbed? In USENIX conference on Operating systems design and implementation
(OSDI) (pp. 1–6).
14. Informa Telecoms & Media. (2013). Mobile SDN: The future is virtual. White paper. http://forums.
juniper.net/jnet/attachments/jnet/IndustrySolutionsEMEA/258/1/Mobile-SDN-The-future-is-virtual-
final%20July. Accessed on 22 February 2013.
15. C. M. R. Institute. (2011). C-RAN: The road towards green RAN. http://labs.chinamobile.com/cran.
Accessed 22 February 2013.
16. Chaudet, C., & Haddad, Y. (2013). Wireless software defined networks: Challenges and opportunities.
In The international IEEE Conference on microwaves, communications, antennas and electronic systems
(COMCAS) (pp. 1–5).
17. Qadir, J., Ahmed, N., & Ahad, N. (2013). Building programmable wireless networks: An architectural
survey. CoRR abs/1310.0251.

Slavica Tomovic graduated in Telecommunication Engineering in

2013 at the University of Montenegro. She is currently working on
her MSc thesis at the Faculty of Electrical Engineering, University of
Montenegro in Podgorica. Her main research interests are in the areas
of software-defined networking, quality of service (QoS) management
and architectures, Internet of things (IoT) and 5G wireless network
design.

123
1644 S. Tomovic et al.

Milica Pejanovic-Djurisic is a full professor in telecommunications

at the University of Montenegro, Faculty of Electrical Engineering,
Podgorica, Montenegro. She is also a director of the Research Cen-
ter for Info-Communication Technologies at the University of Mon-
tenegro. Mrs. Pejanovic-Djurisic has got PhD degree in Telecommu-
nications at the University of Belgrade. She has been teaching at the
University of Montenegro basic telecommunications courses on grad-
uate and postgraduate levels, as well as courses in mobile communi-
cations and computer communications and networks, being the author
of four books and many strategic studies. She has published more than
200 scientific papers in peer-reviewed international and domestic jour-
nals and conference proceedings. Her main research interests are: wire-
less communications, 5G wireless networks, cooperative and energy
efficient transmission techniques, optimization of telecommunication
development policy. Prof. Pejanovic-Djurisic has considerable indus-
try and operating experiences working as industry consultant (Ericsson,
Siemens) and Telecom Montenegro Chairman of the Board. She has been in charge of wireless networks
design and implementation in Montenegro and in the region of SEEurope. Prof. Pejanovic-Djurisic has been
leading and coordinating many internationally and EU funded ICT projects with domestic and international
partners. She is an evaluation expert for EU FP7, and now HORIZON 2020, projects in the field of ICT
and a member of IEEE and IEICE. Prof. Pejanovic-Djurisic has also a considerable experience in the field
of telecommunication regulation. Being an ITU expert, she participates in a number of missions and ITU
activities related with regulation issues, development strategies and new technological solutions.

Prof. dr Igor Radusinovic received the Telecommunications Engi-

neering degree by the University of Montenegro in 1994. He received
the MSc and Ph.D. degree by the University of Belgrade, Serbia, in
1997 and 2003 respectively. He is a Full professor in Telecommu-
nications networks and Switching systems at the Faculty of Electri-
cal Engineering, University of Montenegro. From 2009 to 2011 Prof.
Radusinovic was Deputy Minister for Science, Research and Techno-
logical Development in the Ministry of Education and Science of the
Government of Montenegro. He was Montenegrin representative to the
Board of Governors of the Joint Research Centre of the European Com-
mission (JRC EC) and European Research Area Committee (ERAC).
Prof. Radusinovic was Chairmen of Montenegrin Council for Science
and Research. He is a Member of the Managing Board of University
of Montenegro and President of the Managing Board of. Montenegro
Post He has published as an author or co-author more than 120 papers
in international and national scientific journals and international and
regional conferences. He participated in a number of international (FP7, COST action) and national research
teams and projects, as well as bilateral research cooperation. Prof. Radusinovic has considerable industry and
operating experiences working on many strategic studies, regulatory issues, development strategies and tech-
nical solutions in area of ICT. Research interest of Prof. Radusinovic is focused, but not limited, on: packet
switching systems, telecommunications network theory, congestion control algorithms, software defined net-
working and 5G.

123