3220 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO.

4, FOURTH QUARTER 2018

Will SDN Be Part of 5G?

Zainab Zaidi , Senior Member, IEEE, Vasilis Friderikos, Member, IEEE, Zarrar Yousaf , Member, IEEE,
Simon Fletcher, Mischa Dohler , Fellow, IEEE, and Hamid Aghvami, Fellow, IEEE

Abstract—For many, this is no longer a valid question I. I NTRODUCTION

and the case is considered settled with software defined
HILE 5G is still not standardized,1 this paper is try-
networking/network function virtualization (SDN/NFV) provid-
ing the inevitable innovation enablers toward the realization
of a virtualized, programmable, and flexible 5G network. As
W ing to point out that there is a significant chance that
the idea of a fully programmable control plane with switches
SDN along with other technology enablers (including NFV) are may not be tech-ready for 5G. The emphasis of this paper is
still in the process of evolution, the first commercial deploy- only on the mobile or cellular networks (both access and core
ment of 5G may take few years. However, some companies
are claiming the availability of 5G solutions, but consider-
networks), although, 5G may encompass several other tech-
ing the monumental task of softwarization of mobile cellular nologies for connectivity, i.e., satellite, cable, optical, WLAN
networks there are genuine concerns that we may only see (Wireless LAN) etc.
some point solutions involving SDN technology instead of a The major focus of SDN technology is towards decou-
fully virtualized SDN-enabled 5G mobile network. In order pling of the software-based control plane from the hardware-
to determine the technology readiness of SDN solutions in
the context of 5G networks, this survey paper attempts to
based data plane (e.g., packets forwarding) of networking
identify all important obstacles in the way, and looks at and switching pieces of equipment. SDN brings with it
the state of the art of the relevant research. This survey unprecedented ease in innovation, openness, optimum resource
is different from the previous surveys on SDN-based mobile utilization, support for virtualization, etc., but it is perhaps
networks as it focuses on the salient problems and discusses the most disruptive idea, the mobile networks have seen since
solutions proposed within and outside SDN literature. Our
main focus is on fronthaul, backward compatibility, suppos-
their transition from 1G to 2G, i.e., from analog to digital
edly disruptive nature of SDN deployment, business cases and systems. Even LTE did not come with the similar degree of
monetization of SDN related upgrades, latency of general pur- disruption [1]. The disruption required may possibly push the
pose processors, and additional security vulnerabilities that SDN deployments, and as a consequence the development of
softwarization brings along to the mobile network. We have SDN-based solutions, to further future. 5G may just see some
also provided a summary of the architectural developments
in SDN-based mobile network landscape, including deployment
point solutions with SDN in some places [1] and the focus
options for SDN within NFV framework, as not all work would largely remain on densification, mm-wave technology,
can be covered under the focused issues. This paper provides and DAS (Distributed Antenna Systems) to achieve 1000x
a comprehensive survey on the state of the art of SDN- capacity, 100x data rate, and 100x active connections from
based mobile networks and clearly points out the gaps in the the already deployed LTE network. An interesting point here is
technology.
that all of the salient targets and business opportunities defined
Index Terms—5G mobile/cellular network, radio access for 5G are not strictly dependent on SDN. Nevertheless, it is
network (RAN), SDN, NFV, control plane, data/forwarding important to mention that 3GPP’s 5G NR (New Radio) is a
plane, OpenFlow, CloudRAN, LTE, EPC, fronthaul, backward
compatibility. step closer to SDN-based mobile networks due to complete
separation of control plane (CP) from user plane in NGC (New
Generation Core), modularizing of CP functions, and defining
the interfaces between CP functions as services [2].
In recent years, there has been a lot of interest in SDN-based
mobile network and there are a number of papers proposing
Manuscript received August 12, 2017; revised January 21, 2018 and March SDN-based mobile networks architectures and listing the huge
23, 2018; accepted May 3, 2018. Date of publication May 21, 2018; date
of current version November 19, 2018. This work was supported by the benefits they can bring to the mobile industry [3]–[11]. These
EC H2020-ICT-2014-2 under Project 5G NORMA. (Corresponding author: papers were basically advocating the case for SDN by empha-
Zainab Zaidi.) sizing the positives, a very understandable trend to address
Z. Zaidi is an Independent Researcher, Ilford, U.K. (e-mail:
zzaidi@ieee.org). the need of the time. As the time passes, researchers are
V. Friderikos, M. Dohler, and H. Aghvami are with the Centre for also raising concerns about getting the technology ready for
Telecommunications Research, King’s College London, London WC2R deployments and respective obstacles along the road. The state
2LS, U.K. (e-mail: vasilis.friderikos@kcl.ac.uk; mischa.dohler@kcl.ac.uk;
hamid.aghvami@kcl.ac.uk). of the art of SDN technology in wired and fixed network is
Z. Yousaf is with the Department of Mobile Networks and Systems
Group, NEC Laboratories Europe, 69115 Heidelberg, Germany (e-mail: 1 The Non-standalone 5G NR (New Radio) specifications are recently
zarrar.yousaf@neclab.eu). approved in Dec. 2017 and the rest of Rel. 15 (Standalone 5G NR) has
S. Fletcher is with Real Wireless Ltd., Pulborough RH20 4XB, U.K. expected functional freeze date of Sep. 2018. The work on standardization
(e-mail: simon.fletcher@realwireless.biz). of NFV has moved already to detailed specification and proof-of-concept by
Digital Object Identifier 10.1109/COMST.2018.2836315 ETSI. IEEE is also developing standard for NGFI, a new fronthaul technology.
1553-877X  c 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
ZAIDI et al.: WILL SDN BE PART OF 5G? 3221

much more advanced than the SDN-based mobile networks As per our knowledge, none of the surveys on SDN-based
developments. However, with 87% of the total Internet users mobile networks has looked into the matter of obstacles as the
now in possession of a smart-phone,2 mobility is very impor- main focus. In Section V, we have presented a brief account
tant for global telecommunication network. LTE Evolved of available surveys on SDN-based mobile networks proposals
Packet Core (EPC) has done a remarkable job in simpli- and how our paper is different from them.
fying the core and separating control and user plane to an Once again, we want to emphasize that this paper is not
extent. 3GPP’s 5G NGC has further improved this separa- disputing the benefits of SDN rather we consider them substan-
tion. The base-station eNodeB, however, still contains both tial and necessary for the future generations of networks. Our
planes. Moreover, shared wireless medium and interference question is not “why?”, it is “when?”. Considering the gigantic
between neighboring base-station make it more difficult to task, it is vital to prioritize the research issues in order of their
realize separation of control from forwarding plane. impact over the SDN technology’s development and realiza-
We have identified six major obstacles and issues which tion. The main contribution of this survey paper is to provide a
need to be addressed in order for the technology to move clearer picture to the readers about what is needed to be done
forward. There are certainly other issues as well and as regarding the realization of SDN-based mobile networks. We
the technological development in this area progresses, there achieved the objective by:
will be an even longer list of questions. We have selected 1) identifying the major issues and obstacles which should
the following six issues according to their relevance with be resolved so the technology could reach the deploy-
research and innovation. Other issues, such as, re-organization ment stage;
of telecommunication industry to suit the SDN model, regu- 2) providing the state of the respective solutions, lessons
latory issues, etc., are outside the scope of the present paper. learned, comparison of alternate proposals wherever pos-
Nonetheless, they are equally important issues to be discussed sible, and highlighting the gaps through summaries,
by the SDN community. Moreover, other issues related to tables and lists. In particular
optimization, added functionalities, etc., are also outside the a) fronthaul we compared different functional splits,
scope of present paper. Examples of such issues are energy different transport protocols, such as, Ethernet or
efficiency of SDN networks, optimized resources consump- CPRI, and different hardware options, such as,
tion, intelligent networking techniques for big data, integration fibre or microwave;
of IoT (Internet of Things) [12], network architecture for rural b) latency of GPP we compared available LTE imple-
and low-income areas [13], etc. However, these issues are also mentations in GPP and it is very clear that more
important, e.g., the additional energy and resource consump- work is required to be done here;
tion introduced by SDN and NFV should also be considered c) backward compatibility we explored different
and efficient techniques should be developed for greening the definitions of the issue and examined the respective
future networks [12]. The major obstacles are: state of the art;
1) fronthaul; d) deployment we discussed SDN deployment
2) latency of general purpose platforms; options, i.e., revolutionary or evolutionary, white
3) backward compatibility; box or overlay solutions, etc., and respective avail-
4) disruptive deployment; able solutions;
5) SDN specific security vulnerabilities; e) security we presented a comprehensive threat list
6) clear and compelling business case. and possible solutions grouped under the associ-
We explain the issues in detail in Section III and they set the ated SDN layer; we also discussed summary of
standard according to which we survey the existing proposals wireless specific security solutions;
of SDN-based mobile networks. It is very interesting to see f) business case we summarized results from cost
that although there are many publications regarding the new analysis studies to show the expected cost savings
architecture, very few touch upon the real hurdles. Although, from softwarization, cloudification, and virtualiza-
there has been significant efforts to explore solutions for the tion of network; we created a list of commercial
above mentioned issues but they are sporadic and isolated and products for SDN-based mobile networks;
a comprehensive solution is still missing. The main motivation 3) pointing out the standardization activities wherever pos-
behind this survey paper is to understand the state of the art sible;
and highlight the gap between what is being done versus what 4) surveying the architectural development in SDN-based
is needed to be done. A big picture or broad perspective is a mobile networks around the popular themes to get an
compass to help us steer towards the goal. Although, it is not understanding of the outlook of future networks.
trivial to have a big picture for such a complex problem show- The paper is organized as follows: Section II contains a
ing all significant dimensions of the problem space. Major brief background about SDN, NFV, CloudRAN, and current
obstacles and issues in the path of technology deployment is and future LTE architectures. This section is added to make
one way of constructing the big picture which is meaningful this paper self-contained, readers knowledgeable in the above
and very effective in understanding the state of the art. technologies can easily skip this section and go directly to
Section III. As discussed above, Section III discusses the
2 http://www.globalwebindex.net/blog/87-of-internet-users-now-have-a- major obstacles and issues in detail and also lists the solu-
smartphone tions proposed in the literature. Section IV then summarizes
3222 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

Fig. 1. A generic SDN model of three layers.

Fig. 2. The fail-fast model, major benefit of SDN technology.

the SDN-based mobile networks literature as the current state

Currently, any upgrade in mobile network takes a long time.
of the art depicting specially the achievements unrelated to
With SDN, any mobile technology will have much smaller
the obstacles and issues identified in this paper. Section V
trial and deployment process.
discusses the SDN-based mobile networks’ surveys and what
The LTE architecture also distinguishes between user-plane
was the need to have the present one. Finally Section VI
and control-plane, where the former deals with data packet for-
summarizes the major take-away points made in this paper.
warding, while the latter focuses on signaling and management
messages and operations using the same physical infrastruc-
II. BACKGROUND ture. Both planes reside in the firmware of the system. This
In this section, we present brief discussions about SDN, demarcation is much clearly designed into the mobile core
NFV, CloudRAN and LTE including some future proposals network EPC (Evolved Packet Core), but the radio access part
for mobile network architecture. of LTE consists of only base-station node, eNodeB, performing
data forwarding and control functions. The SDN-based core
A. SDN network requires transporting the control plane into software
along with the control logic required for the data forward-
SDN mainly focuses towards decoupling of the software-
ing plane, e.g., routing rules, mobility anchoring, etc. We also
based control plane from the hardware-based data plane (e.g.,
remark that user plane as defined by LTE is not exactly the
packets forwarding) of networking and switching pieces of
same as data forwarding plane of SDN and similarly both
equipment [14]. The basic SDN model is shown in Fig. 1.
control planes also differ slightly [15].
The logically centralized controllers contain the control logic
to translate the application requirements down to the data plane
and are responsible in providing an abstract network view to B. NFV
the application plane. According to [14], SDN is solution to In NFV (Network Function Virtualization), network func-
manage the complex and dense 5G network with many servers tions are implemented in software components called virtual
and routers. The major issue is to create appropriate mapping network functions (VNFs) [16]. VNFs are deployed on com-
of the existing network functionalities to the decoupled control modity servers or cloud infrastructure instead of dedicated
and forwarding planes. Conforming to the terminologies used hardware reducing CAPEX (CAPtial EXpenditure) and opti-
in the research community, the interface south of control plane mizing resource provisioning [16]. The VNFs might not only
is called the southbound interface in Fig. 1 and north of control be linked to a specific service and/or application but they
plane is the northbound interface. An important point to note can relate to specific user requirements, network policies and
here is that both interfaces define the bi-directional information allow for service differentiation based on the user subscription
flow contrary to what may be implied by their names. SLAs (Service Level Agreements). The key motivation is that
The major benefit of softwarization is flexible innovation as once implemented in software these VNFs are able to run in
other advantages come indirectly, such as, efficient resource general-purpose hardware where they can be deployed, con-
utilization is a result of centralization not softwarization. Also, figured, scaled, migrated, and updated on demand. Operation
novel business opportunities through RAN-as-a-service (RaaS) of the VNFs is expected to take place in accordance to the
is a possibility through cloud computing and virtualization ETSI Network Function Virtualization (NFV) management
which not necessarily need softwarization. Similarly, NFV is and organization (MANO) framework depicted in Fig. 3 [17].
a complementary technology to SDN and does not depend on The MANO framework defines three main logical compo-
it [1]. Flexible innovation or fail-fast approach, as shown in nents, namely the Virtualized Infrastructure Manager (VIM),
Fig. 2, is the important feature which is not possible without the VNF Manager (VNFM) and the NFV Orchestrator
SDN. In fail-fast model, ideas are tried through rapid deploy- (NFVO), interconnected over specific reference points. The
ment, discarded if failed, and scaled out quickly if taken off. VIM manages the infrastructure resources (e.g., compute,
ZAIDI et al.: WILL SDN BE PART OF 5G? 3223

Fig. 4. LTE, RAN and core (EPC) network. The colors indicate the SDN
functional decomposition: orange color shows application plane, green refers
to control plane functions, and turquoise shows data plane. Solid lines show
wired connections.

Fig. 3. ETSI NFV Management and Orchestration (MANO) Framework is clear from Fig. 4, the traditional networks have control and
Overview. forwarding planes integrated into all elements and a proper
logical decomposition, while preserving the performance con-
straints over the inter-communication between controller and
network, storage, memory) whereas the VNFM performs data plane elements, is the biggest challenge.
the life-cycle management of the individual VNFs (such as In this section, we also discuss some of the ideas proposed
scaling, migrating, update/upgrade etc). The NFVO is respon- for future mobile network architecture, i.e., Beyond Cellular
sible for both the resource/service orchestration of a Network Green Generation (BCG2) [19] and Phantom cell concept [20].
Service (NS), which is composed of one or more VNFs (and These ideas have shown big promises in energy efficiency [21].
its components) interconnected over virtual links. There are The logical decoupling of data transmissions and control sig-
additional data repositories that may contain necessary infor- naling paradigm is one of the key directions being explored
mation about NS, VNF, NFV and NFVI (NFV Infrastructure) by GreenTouch3 under the project Beyond Cellular Green
that will enable the NFVO to perform its tasks. The MANO Generation (BCG2) [19]. In BCG2 architecture, the signal-
architecture also defines reference points for interfacing the ing nodes are responsible for the coverage and are usually
MANO system with external entities like NFVI, OSS/BSS, assumed to deliver low rate services, such as, random access
VNFs and Element Managers (EM) for delivering a unified and paging, over long ranges; whereas the data nodes can be
management and orchestration of a VNF system. activated and deactivated depending on the traffic demand and
SDN is a complementary technology to NFV and the it is designed for high rate and small ranges. The decoupling
realization of NFV does not depend on SDN. is logical in nature and a single location can host both types
of nodes. A set of studies regarding the BCG2 architecture
C. LTE Architectures- Current and Future is performed under the EU FP7 IP project EARTH [21]. The
The LTE architecture [18], shown in Fig. 4, also distin- study shows that up to 85-90% saving potential is possible
guishes user plane, dealing with the data packet forwarding, with this revolutionary changed architecture compared to the
and control plane, focusing on signaling and management current systems. Coverage is also separated from data process-
messages and operations, using the same physical infrastruc- ing in a CloudRAN architecture [5], [22], where a centralized
ture. EPC (Evolved Packet Core) has the designated control BBU pool serves several RRH (Radio Remote Heads) in the
plane nodes MME (Mobility Management Entity), HSS (Home area and not-in-service BBUs can be put in sleep mode to
Subscriber Server), and PCRF (Policy Control and Charging save energy. Energy efficiency is also discussed extensively
Rules Function) [18]. Serving gateway (S-GW) and packet with in the context of cellular and cloud networking in terms
gateway (P-GW) are used for the data forwarding. From of optimization of transmission power control, sleep cycles
SDN’s perspective, the gateways and the base-station also for infrastructure resources, etc. [23], [24]. For IoT (Internet
contain control plane functionality along with the designated of Things) and D2D communications, the issue of energy effi-
control nodes of EPC. This is shown by color coding all EPC ciency also encompasses load balancing and enhancement of
nodes and eNBs (eNodeB, base-station) partially with green battery life [25].
color in Fig. 4 relating them to the control plane in Fig. 1. The Phantom cell concept [20] which was introduced for real-
base-stations eNBs and S/P-GWs are comprised of the data izing true potential of dense deployment of small cells as
plane from SDN’s point of view and thus shown with partial suggested for LTE Release 12. In this idea, many small
turquoise coloring in Fig. 4. A possible implementation of EPC cells, called the phantom cells as they contain only LTE
control nodes in a SDN environment is as application modules user plane, are overlaid with a normal macro cell which
so we color them partially orange. The colors are relating the
nodes in Fig. 4 to the layers in SDN model of Fig. 1. As it 3 www.greentouch.org
3224 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

provides interference coordination. However, a macro cell con-

sumes over 100 times more power than a pico/femto cell as
described in the EU FP7 EARTH project [21] and keeping
it on all the time would lead to severe power inefficiency.
Reference [15] proposed to change the macro base-station in
Phantom cell architecture with a signaling only large cov-
erage base-station which can also hold some control plane
processing for SDN-based mobile network. The architecture
can provide interference coordination service to the small
base-stations in the area.
Although not yet explored, but the phantom cell or the
data cell in BCG2 architecture can become a starting point
to realize dump base-stations in a SDN-based mobile network
architecture. The dump and small base-stations can provide
Fig. 5. The CloudRAN architecture, with BBU pool and EPC in the cloud.
data forwarding only and their control lies elsewhere, e.g., Colors indicate the SDN functional decomposition: orange shows application
in the cloud or in the macro base-station. Striping down sig- plane, green refers to control plane, and turquoise shows data plane. Solid
naling capability from dump base-stations opens up bigger lines show wired connections.
challenges as how to determine the best small base-station to
connect to UE. Moreover, a full SDN-based mobile network
can also allow us to choose the base-stations to remain on for the load on the fronthaul by leaving some processing blocks
signaling instead of a predetermined signaling base-station in at the base-station [8]. Also, some proposals have left the
BCG2. In a CloudRAN type architecture, the controller can part of control plane at the base-station in order to perform
select a set of RRHs to provide signaling functions and the delay-critical control [3]. These proposals require connections
rest RRHs can be used on-demand. between control elements in the cloud and at the base-stations.
The CloudRAN architecture has the advantage of lower
D. CloudRAN cost, energy efficiency, smaller footprint, and lower main-
tenance cost [27]. Alcatel Lucent’s Light Radio, Ericsson’s
CloudRAN has its roots firmly planted by China Mobile4
Antenna Integrated Radio and Nokia Siemens Networks’
and it can be viewed as a facilitator for introducing SDN tech-
Liquid Radio are among the earlier CloudRAN products.
nology. CloudRAN, which is now hitting its stride, centralizes
The CloudRAN architecture, with its low cost and flexibility
all functionalities, control as well as data plane, into a central-
of operation, also opens up new possibilities. The architecture
ized BBU (Base Band Unit) pool, or a data center/cloud, for
is shown to be extremely useful for very high-speed train com-
easier management and coordination as shown in Fig. 5. In that
munications5 [28]. Due to low cost, several remote radio heads
respect, only antennas and some active RF components, i.e.,
(RRHs) can be deployed along the track in the framework
RRU (Remote Radio Unit) or RRH (Remote Radio Heads),
presented in [28]. Centralized processing allows to use virtu-
are left on the cell sites [5], [22]. Figure 4 is morphed into
alization along with joint beamforming techniques to yield up
CloudRAN architecture by replacing eNB with RRHs/RRUs
to 100Mbps at the speed of 450kmph [28].
and moving the processing units, i.e., BBUs and the EPC to
the cloud as shown in Fig. 5. CloudRAN also does not decom-
pose control and forwarding layers as shown by the color III. M AJOR O BSTACLES AND I SSUES
coding in Fig. 5 to match with SDN layers of Fig. 1. Although,
the RRHs/RRUs can be classified as the data forwarding RF We identify six major issues hindering the realization of
devices without any control functionality. SDN technology. In no particular order, these issues are
CloudRAN is proposed as a mechanism to realize small 1) Fronthaul
cell deployment in LTE through proper coordination for 2) Latency of general purpose platforms
interference management [26]. CoMP (Coordinated MultiPoint 3) Backward compatibility
or orthogonal/non-orthogonal random access are typical meth- 4) Disruptive deployment
ods for interference management [26]. LTE small cell, how- 5) SDN specific security vulnerabilities
ever, assumes distributed control with self-organizing (SON) 6) Clear and compelling business case
capabilities. The connection from the cell site to the serving In this section these issues are discussed in detail. We have also
gateway (S-GW) in LTE could be through wired or wireless looked into the state of the solutions proposed for them. This
links (http://scf.io/). On the other hand, CloudRAN assumes is one of the most rapidly changing field in digital technology
a high bandwidth fiber link between RRU and the centralized and last few years have seen massive progress in the area.
data center, which is also the most prohibitively expensive We cannot claim to have looked at each and every publication
aspect of this proposal. under the relevant themes but we have certainly covered the
Centralized processing in the cloud is not necessary for major solutions and ideas presented in the literature including
SDN and alternative proposals have been proposed to reduce
5 https://www.globalrailnews.com/2017/12/04/samsung-successfully-tests-
4 http://labs.chinamobile.com/cran 5g-on-high-speed-train-in-japan/
ZAIDI et al.: WILL SDN BE PART OF 5G? 3225

industry white papers and technical reports as well as academic

research papers.

A. Fronthaul
In centralized RAN architectures, such as, CloudRAN, con-
nections from remote radio units (RRU) to the centralized
baseband processing units (BBU) form fronthaul as opposed
to backhaul, which connects the base-stations to the core
network. RRUs are the wireless transceivers mainly consisting
of the antenna heads.
Traditionally, data from the processing cabinets of base-
stations, in the form of digitized baseband signal, is carried to
Fig. 6. Fronthaul functional split options (courtesy [22]).
the antenna heads located at the rooftop or top of the masts
by CPRI (Common Public Radio Interface) protocol over fiber
optic connections. CPRI over fiber allowed the antennas to
be located away from the base-station cabinets as the legacy In fog/edge computing architectures [34], the nodes are
co-axial cables caused a lot of power loss and were also widely spread, away from the cloud or data center, where the
expensive and bulky [29] and can only be used for short dis- users are. They can provide some localized services and can
tances. CPRI over fiber can connect RRUs miles away from also store some information. Such architecture also reduces
BBUs with minimal losses, however, the limiting factor is the the load on fronthaul with some improvement in latency
latency constraint to maintain channel state information and requirements.
synchronization between BBU and RRU [29]. An effort to move away from semi-proprietary CPRI has
The data to and from the centralized BBUs is digital ver- also been made by ETSI through their Open Radio Equipment
sion of high frequency baseband signal. As an example, a LTE Interface (ORI) specifications.6 The ORI interface was built on
baseband signal with 20MHz channel bandwidth needs to be top of CPRI and it supports line bit rates up to 10.14Gbps.
sampled 30.72 million times per second for proper digitiza- The objective was to make the interface fully inter-operable
tion resulting in data rate of approximately 2.5Gbps if data and improving the data transport efficiency was not the main
stream is duplicated for 2 × 2 MIMO [29]. CPRI rate ranges goal.
from 614Mbps to 12Gbps [29]. Such high data rates need fiber Among the various SDN-based mobile network architec-
optic cables, although, microwave can be used but for lower tures proposed in the literature, some have tried to touch upon
rate CPRI only. Interestingly, CPRI is very inefficient from fronthaul but not really looking into resolving the transport
data transport point of view, e.g., a 20MHz LTE channel can efficiency or latency issues. As shown in this section, the fron-
carry up to 150Mbps in downlink but requires CPRI rate of thaul is a big research problem in itself and we should not
2.5Gbps [29]. CPRI was not designed to carry data over very expect that research efforts towards new architectural organi-
long distances and it was an internal base-station interface to zations for SDN-based mobile networks would also be able
connect the antennas on the rooftop to the base-station cab- to resolve fronthaul issues as well. But since fronthaul is a
inet. There is clearly room for introducing efficiency in data major element of SDN-based mobile network architectures,
transport in fronthaul. Advanced techniques exploiting corre- it cannot be left behind. Interestingly, we found only two
lation among RRUs, such as, data compression, quantization, papers [3], [8] which addressed the issue in their architec-
precoding, etc., can be used as well to reduce the data rate for tural design. Details about the design is discussed later in
the constraint fronthaul [30]. Section III-A1. The only other papers which discussed fron-
Other important issues for fronthaul are latency and jitter. thaul in their architecture is CONCERT [11] and SDCN [9]
Real-time communication, such as, VoLTE (voice over LTE), but the major relevant issues were not their focus. CONCERT
IoT, tactile applications, etc., and the control signaling between and SDCN proposed that fronthaul network can also be a SDN
RRH and BBU to maintain channel state information require network with optical switches controlled and virtualized by
low delays. To keep one way delay under 75 microseconds, the SDN controller. CONCERT claimed virtualization efficiency
length of the fiber link between RRH and BBU should not be with SDN fronthaul than using packet switched Ethernet [11].
more than 15Km [29]. According to [31], the delay of CPRI 1) Functional Split Between BBU and RRU: Different cen-
should remain under 100µs and jitter should be under 65ns. tralization options are also considered to reduce the data rate
Jitter in the synchronization signal between BBU and RRU for fronthaul [22], [30] as shown in Fig. 6. When process-
will induce phase noise which in turn will lead to degrada- ing for all layers is performed in the centralized BBUs, it is
tion in the transmitted modulated signal [32]. Heterogeneous called full centralization as shown by option 1 in Fig. 6. In
CloudRAN [33] tried to solve the latency issue by introduc- this case, we have very high data rate waveform to transmit
ing a different high power node (HPN) than RRUs which to and from the RRUs. Partial centralization options are also
is responsible for control signaling in a geographical area. available where layer 1 or PHY was kept at the RRUs as it
However, the latency issue with data transmissions, such as,
VoLTE, IoT, and other time-critical applications remains. 6 http://www.etsi.org/technologies-clusters/technologies/ori
3226 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

reduces a lot of fronthaul overhead [30]. This is shown as SoftAir [8] is one of the two research papers on SDN-based
option 2 in Fig. 6. mobile network architecture which also addressed fronthaul
Moreover, it sounds desirable to have all programmable OSI transport efficiency issue. In SoftAir modem (modulation and
layers but there is a strong consensus among network owners demodulation) is recommended to be kept at RRU in order to
and operators not to allow low level network programming reduce the data rate of CPRI supported fronthaul. Such a split
by third parties [1] due to network security and management is also evaluated through hardware testbed experimentations
issues. Also, keeping layer 1 on dedicated hardware instead in [38] and latency and jitter are found to be within acceptable
of general purpose processors can also improve the latency bound. The fronthaul is using Ethernet technology, although,
issue which is very critical for some IoT, tactical and real time the length of the fronthaul is not reported in [38].
applications. On the other hand, efficient implementation of SoftRAN [3] provides another type of functional split, in
some advanced features, such as, CoMP (Coordinated Multiple terms of control functions, between center and remote site.
Point) and cooperative processing for massive MIMO will not They also realized that the biggest challenge in realizing
be possible if layer 1 remains in the hardware. Interaction SDN-based mobile network is the inherent delay between any
between layer 1 and 2 will also be complex and one way centralized controller and the radio elements. The delay could
latency between RRU and BBU should be of the order of 100- typically be 5-10ms. SoftRAN splits the control plane func-
150 microseconds [22]. Moreover, the fronthaul should also tionalities and leaves the time-critical local control related
cater for some control signaling regarding carrier frequency, to channel state, such as, downlink resource allocation (with
transmitter power, etc. transmit power set by the centralized controller), at the
Another option for partial centralization is to split MAC individual radio element.
function between the cloud and the antenna site. The one-way A detailed evaluation of different functional splits is pro-
latency constraint can be more than a millisecond. This alter- vided in [39] in terms of multiplexing gains. The obvious
native is shown by option 3 in Fig. 6. In this architecture, conclusion is the best multiplexing gain for full centraliza-
HARQ of MAC is kept at RRU whereas the MAC scheduler tion solution with all processing in the BBU pool. This is
resides in the central location [22]. In another option, only specially true for high load. When the traffic is low, the
layer 3 resides in the cloud and PHY and MAC are kept at fronthaul requirements are shown to be relaxed with only
the antenna site [22]. This also reduces a lot of fronthaul over- higher layers in the BBU pool and lower layer processing
head and handover overhead is also greatly reduced in case near the RRU [39]. The impact of functional split of fron-
of mobility within the serving area of layer 3 as all cells can thaul on its performance is studied with packetization and
be aggregated and provided a single S1 view towards core scheduling in [40] and [41] and it is shown that there is a
network [22]. strong interplay between packetization overhead and latency
Chinamobile presented NGFI (Next Generation Fronthaul that changes the fronthaul performance. For each split, the
Interface) [35] which decouples the dependency of CPRI on analysis finds the maximum number of supported RRUs for
number of antenna elements by putting all antenna related the best packetization method satisfying fronthaul capacity
functionality, such as, downlink antenna mapping, fast Fourier and latency constraints [40]. This work exploits packetiza-
transform [FFT], channel estimation, and equalization in RRU. tion latency and options of putting more processing at RRU
It is shown that an LTE fronthaul bandwidth may decrease to support more RRUs in the system for specific fronthaul
on the order of 100Mbps no matter how many antennas are constraints. The follow-up paper [41] also includes different
used [35]. Currently, NGFI is being standardized under IEEE scheduling mechanisms in the study to improve the number
1914.1 project. of supported RRUs.
Similar ideas of doing cell based processing, such as, A summary of split functionality proposals are presented
FFT/IFFT, subcarrier mapping, etc., of the receiving waveform in the Table I, where it is clear that transport efficiency can
in RRUs were discussed in [36]. According to the authors, be improved with proper distributions of functions between
this split will half the fronthaul capacity requirement and fur- BBUs and RRUs. Although, no study has been done to com-
ther reduction would be possible by moving demodulation and pare the different proposals and we have used the generic terms
decoding blocks to the RRU. They also suggested to put full ‘Heavy’ and ‘Light’ to show that all the split function propos-
PHY for transmitting signal in the RRU to reduce fronthaul als have lesser load than CloudRAN. It is not known how
capacity. they compete with each other. In Table I, RH stands for radio
Similar ideas of performing FFT/IFFT near the users is dis- heads or transceivers and RRC is Radio Resource Control.
cussed and evaluated in [37], where LTE and DOCSIS (Data Also, SoftAir [8] and NGFI [35] did not mention any delay
Over Cable Service Interface Specification) share the fibre figure explicitly and we have taken ≤75µs as it is considered
fronthaul and remote FFT node. The remote FFT/IFFT reduces as a standard for centralized RAN [29]. The split architec-
the fronthaul bit rate to approximately 1/30th of the conven- tures of SoftAir and NGFI focus on only reducing the load of
tional baseband approach. The system also used frequency fronthaul. In SoftAir, delay is considered as a constraint while
domain I/Q symbols instead of conventional time domain I/Q forming clusters of RRUs.
symbols to reduce the required transport bit rate and used 2) Ethernet vs. CPRI/ORI: The use of Ethernet for fronthaul
caching of the repetitive QAM symbols to further reduce the is also been looked at with varius options of placing CPRI
downstream bit rate requirement. Evaluations show 6.5% (for data or RF data directly into Ethernet frames [32]. Ethernet
full load) to 41% (for 10% load) savings for LTE traffic. is a mature technology with standardized OAM (Operations,
ZAIDI et al.: WILL SDN BE PART OF 5G? 3227

TABLE I
S UMMARY OF S PLIT F UNCTIONS

Administration, and Management) capabilities but by putting A significant advantage, in the case of backhaul, is the use
the data in Ethernet frame, the synchronization information of Ethernet instead of CPRI which has much better data trans-
could be lost [32]. Synchronous Ethernet, where clock infor- port efficiency. Moreover, commodity, or low-cost industry
mation is extracted from the received data, and Precision Time standard equipment can be used lowering the overall CAPEX
Protocol, where time-stamped packets are exchanged, can be where fiber is unavailable.
used to achieve synchronization in fronthaul Ethernet operation. The most important development in this regard is IEEE
Moreover, the traditional Ethernet switching cannot fulfill the 1914 working group (NGFI). The working group has two
strict timing requirements of fronthaul traffic and substantial active projects: IEEE 1914.1 is studying NGFI of Chinamobile
optimization to the switching fabric is required [42]. and IEEE 1914.3 is looking at encapsulation of digitized
IEEE 802.1CM Time-Sensitive Network task group is look- radio signal (I/Q samples) into Ethernet frames for fron-
ing into the set of standards focusing on time-synchronized thaul transmission. Also called as ’Radio over Ethernet’, IEEE
low latency high bandwidth services over Ethernet networks 1914.3 consolidates the efforts of a previous working group
for fronthaul transport network. In a centralized RAN, strict IEEE 1904. The Ethernet frame structure and MAC remains
requirements are defined between the RRU and the BBU, unchanged under this project and the focus is on encapsula-
such as, less than 0.1ms one-way latency at 614.4 to tion and mapping of digitized payload as well as control and
24330.24Mbps [43]. The standardization activity looks into management messages.
techniques to reduce latency and jitter, such as prioritizing An mmWave based fronthaul using Ethernet is presented
the time-sensitive frames, time-synchronization, end-to-end in [46] for heterogeneous CloudRANs. The system contains
resource reservation, etc. small range RRUs and macro RRUs. An Ethernet switch is
In LTE small cell network, a massive coverage of back- added with macro RRU to direct the traffic to the UE or to a
haul is needed instead of fronthaul in centralized RAN and it connected small RRU. In this case, the small RRU does not
is still considered one of the major issues [44]. Both wired need to decapsulate the Ethernet frame as it is done at macro
and wireless backhauls are being used according to the oper- RRU. Preliminary results show that the latency is within the
ating conditions. However, with excessive cost of lying fibre limits for fronthaul operation.
for dense networks and technological and regulatory issues In a recent study, an Ethernet based fronthaul is tested
with millimeter-wave backhauls, it is still an open problem using hardware implementation [38] and latency and jitter are
in small cell network similar to fronthaul in CloudRAN type found to be within acceptable bound. The length of the fron-
access networks. Smart backhaul/fronthaul solutions where thaul is not reported in [38]. Another detailed study in [31]
interworking is exploited and joint optimization is performed claimed that the encapsulation delay would compromise only
with the access network are also recently explored [44]. few kms in fronthaul and it is also possible to reduce jitter,
A futuristic idea of reconfigurable backhaul/fronthaul, even eliminate it, with some scheduling techniques.
namely Xhaul, is presented in [42]. Here functional split 3) Hardware for Fronthaul: The cheapest option for fron-
between remote radio site and centralized processors depends thaul is considered to be WDM (Wave Division Multiplexing)
on the network conditions and operational scenarios. A soft- over dark fiber [1], [29] where the fiber is leased per mile not
ware defined xhaul employs different protocols, such as, by the bandwidth. Places where dark fiber is not available,
CPRI/ORI, Ethernet over heterogeneous physical links, e.g., duct sharing or Physical Infrastructure Access (PIA) can also
fibre, microwave, etc., with appropriate resource allocation to be considered as a solution [1], where cables can be placed in
address latency, jitter, and throughput constraints for each type access ducts using a license agreement. It is very much clear
of data. that countries with fiber rich city infrastructure will benefit far
Another proposal for software defined backhaul is presented more from these savings than those with low fiber availability.
in [45]. Here a SDN orchestrator is responsible to man- Although, fiber deployment is gaining momentum around
age multiple smart-gateways, a new element suggested in the the world but as shown in Fig. 7,7 the growth is mainly in
scheme, in order to provide demand-based uplink capacity to China and Hong Kong with Latin America, Africa, Middle
the connected small cells. The smart-gateways are connected
to the S/P-GW of multiple operators and allow resource 7 Source: http://wireandcablenews.crugroup.com/wireandcablenews/insights/
sharing through SDN orchestrator. free/2015/10/4522446/.
3228 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

scalability issues in centralization of control layer are also

needed be explored in further details besides other technical
challenges [47].

B. Latency of General Purpose Platforms (GPP)

The history of base-station softwarization efforts go back
beyond SDN. Potential cost savings and flexibility in imple-
menting and changing protocols were the main drivers behind
the efforts. Within SDN ecosystem, base-stations are integral
which can have programmable functionalities, i.e., protocols
and transmission parameters, either centralized or distributed,
implemented on a general purpose computing platform and
generic radio heads. By softwarization, the BS and core
network functions can be virtualized thus leveraging on the
Fig. 7. Compound annual growth rate (CAGR) for optical fiber around the advantages offered by virtualization such as flexible and agile
world. deployment and management of network functions and agile
composition of new network functions. In this context, the
performance of GPP became a genuine concern.
East, and South Asia lagging far behind. Satellite connections
GPPs were not designed for real-time signal processing
are suggested as alternatives where governments cannot install
where the delays should be of the order of tens of microsec-
fibre network but cost is still inhibitive [13].
onds [48]. Although, GPPs are also improving their speed by
According to the new statistics released in June 2013,
finding and eliminating the bottlenecks, but it is yet to be seen
fiber-optic links are used for 15.8 percent of broadband con-
if the 1000x capacity and 100x data rate promise of 5G can
nections in the OECD area with Japan and South Korea
be supported by a base-station in GPP. Moreover, the promise
leading the list with 68.5% and 62.8%.8 Organization for
of saving CAPEX in SDN through GPP may not be as huge
Economic Cooperation and Development (OECD) monitors
as the predictions, as the cost of specialized ASIC hardware
economic trends in North America, Europe, Australia, Japan,
is also coming down with time [1]. The actual cost saving
South Korea, Turkey, and some other developed countries and
should also consider this downward trend.
regions. It is very clear that fiber-rich countries will gain the
There has been significant efforts in implementing
real benefit of the SDN-based and centralized RAN but the
LTE eNodeB in software with promising performance
region with low economic incentive to deploy optical network
results [36], [48], [49]. The latency values were shown to be
will be left behind. It is a serious concern that the advance-
decreasing with processor speed. Special techniques are used
ment in telecommunication technology will further widen the
in the software to improve latency using parallelism, replac-
digital divide.
ing computations with lookup tables, kernel patches for time
Other technologies, such as co-axial or twisted pair cop-
guarantees, etc. So far, data rates up to 43.8Mbps on a 20MHz
per cables, microwave, mmWave, are also explored to provide
channel are shown to be supported by a LTE base-station in
fronthaul transport network. In [46], a mmWave fronthaul is
software. It is still a long way to go as far as 5G goals are
used between small RRUs and macro RRUs. Two different
concerned. In this section, we will discuss prominent efforts in
scenarios are tested in preliminary study, i.e., sub-3 GHz and
this respect and assess the gaps in the current state of the art.
Eband (7176, 8186, and 9295 GHz). Eband is showed to have
Reference [48] presents, Sora, a fully programmable soft-
approximately 11% reduced RRUs for the same traffic volume
ware radio platform on GPP from Microsoft. It was the first
due to availability of broader spectrum.
effort to program LTE on a commodity general purpose com-
4) Software Defined Optical Networks: Another important
puter. Sora implemented LTE uplink Rx PHY that supports up
issue in relevance to optical fronthaul is the developments in
to 43.8Mbps data rate on a 20MHz channel. Sora, however,
software defined optical networks so a complete SDN-based
uses some hardware components for high throughput and low
networking solution can be realized. There has been a number
latency data transfer between radio front-end and host memo-
of developments regarding software defined optical infrastruc-
ries [48]. They bridged the synchronous transfer of high data
ture, i.e., transceivers, switches, medium access, performance
rate RF digitized signal and asynchronous processing of a
monitoring, etc. [47]. The survey in [47] provides a com-
GPP through FPGA FIFOs (buffers) and on-board memory
prehensive account of research in control layer framework,
with the support of PCIe bus standard for high throughput
access control, security, QoS monitoring, virtualization mech-
and low latency. They have also provided a separate path
anisms, etc. Despite the developments, important challenges
for low latency control data. They have used new software
are still remaining specially in integration of products from
techniques, such as, use of lookup tables instead of compu-
different vendors and heterogeneous physical layer technolo-
tations, use of special kernel service for core dedication in
gies, e.g., radio and fibre [47]. Security vulnerabilities and
multi-core processor environment, etc., to improve the latency
8 https://www.cnet.com/uk/news/fast-fiber-optic-broadband-spreads-across- for real-time PHY processing. Sora does provide a good start-
developed-world/ ing point but it does not include softwarization of downlink
ZAIDI et al.: WILL SDN BE PART OF 5G? 3229

LTE which is also important and requires more complex FFT Softbox [53] is a recent design of a mobile core network
processing. Also, realizing the projected data rates for 5G, which combines softwarization and virtualization to enable
i.e., in the order of tens of Gbps, it is still a long way to UE-based customized service delivery. Softbox slices the core
go. Moreover, the real-time processing of CoMP (Coordinated resources optimally considering the mobility of UE into var-
Multi Point) interference management scheme to reap the ben- ious UE containers. Proof-of-concept prototype is developed
efits of centralization would be a big challenge as well on using open-source software (e.g., Docker container, RYU SDN
software platforms. controller, OAI EPC). SoftBox is evaluated by combining real
Virtual WiMax base-station pool is developed and evalu- and synthesized LTE traces. Results show that basic SoftBox
ated in [50] which can meet system requirements including has by 86%, 51%, and 83%-87% lower signaling overheads,
synchronization, latency and jitter.The implementation uses a data plane delay, and CPU core usage, respectively, than two
FPGA based adapter board between radio head and virtual EPC systems (i.e., OAI EPC, OpenEPC).
base-station pool implemented on GPP and can support up to srsLTE is an open source implementation of LTE PHY only
10MHz signal bandwidth. on GPP [54] with extensive use of data and instruction paral-
In another effort, reported in [49], LTE eNodeB prototype is lelism and use of different kernels of SIMD (Single Instruction
implemented using single core GPP. The prototype was tested Multiple Data) code for different architectures. One process-
for latency in receiving uplink signals and transmitting down- ing core showed to have met the delay deadlines of LTE but
link signals and the results did not meet LTE standards [49]. multiple cores were recommended to be used for processing
The authors, though, expressed the hope that a multi-core of the rest of the layers. A commercial version of srsLTE is
platform would bring the delays within LTE specifications. also available.
OpenRadio [6] explored the softwarization of PHY layer A LTE eNodeB in software is commercialized by
of a base-station. OpenRadio uses hardware accelerators for Amarisoft,10 where they claim to have a fully LTE release
highly computational blocks, such as convolutional coding, 13 complaint software which can be used with USRP11 radio
Turbo coding, etc. and separates the decision rules from the front ends. The software seems to be meeting all the timing
protocols to be put in the software in order to meet strict requirements but no documentation is available to explain how
deadlines. OpenRadio is also concerned about the software did they achieve it. The company also claims that 500 UEs
based decision plane not to become very heavy as it would can be supported by a single PC but the maximum achievable
incur prohibitive inter-core communication overheads imply- throughput is not mentioned.
ing that decision/processing plane separation might not be the A proprietary virtualized access (vAccess) development
best design choice [6]. platform from Freescale [55] is created to facilitate the devel-
A notable effort in this regard is from EURECOM called opment of SDN/NFV products. It is built over OpenStack
OpenAir Interface.9 OpenAir Interface (OAI) is an open source cloud computing platform with hardware accelerators for L1
implementation of LTE Release 8.6 with parts of Release 10. real-time processing and Linux patch to support bounded guar-
The maximum expected throughput is reported to be 36Mbps antees for application start time. A LTE smart base-station
on a 10Mhz downlink channel. Experiments on OpenAir built over the vAccess platform is also presented in the
Interface show that the BBU processing delay gets lower market.12
with processor speed [36]. A 3GHz single core processor In 2016, NEC announced NFV CloudRAN13 software
is shown to comply with the HARQ delay requirements of which provides flexible functional split between centralized
uplink and downlink processing [36]. However, an independent digital unit and radio unit of a CloudRAN. The software
performance study [51] shows that the downlink execution application runs on GPP platforms but requires hardware
time increases with the load and could reach over 200ms for accelerator for L1 processing. Similarly, ASCOS’s virtual
PDCP payload size of 1500bytes. OpenAir Interface is an on- base-station,14 which came out in 2017, also uses hardware
going project, it is yet to see how the base-station software accelerators for L1 signal processing.
fare for 5G throughputs and data speeds. There are other efforts to implement LTE in open-source
In a recent effort [52], both EPC and eNodeB are deployed software for research testbeds. OpenLTE15 and gr-lte16 are
in one mini Intel commercial PC (i7-5557U). All protocol lay- notable examples but they are not for real-time transmission
ers of eNodeB are implemented on the GPP platform with and reception of LTE signals. gr-lte is only a receiver which
X86 CPU instruction set. The communication between RF is based on GNU radio.
block and BBU block is through high speed USB 3.0 interface. A reconfigurable and programmable SDR platform is
The software based eNodeB has USRP B210 software defined proposed in [56] which uses hybrid FPGA to meet all timing
radio for transmission and reception. Currently, this implemen- requirements instead of GPP. However, it is noted in the paper
tation can support stable real-time signal transmissions with
a bandwidth of 10MHz and a stable data rate of 32Mbps
10 http://www.amarisoft.com/?p=amarilte
to multiple commercial mobile terminals. The authors also
11 https://www.ettus.com/product/category/USRP-Networked-Series
commented about the power consumption of their LTE imple- 12 http://sooktha.com/sbs.html
mentation which is comparatively more than the dedicated 13 https://www.sdxcentral.com/products/nfv-c-ran/
hardware. 14 https://www.sdxcentral.com/products/asocs-virtual-base-station-vbs/
15 https://sourceforge.net/projects/openlte/
9 http://www.openairinterface.org/ 16 https://github.com/kit-cel/gr-lte
3230 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

TABLE II
S UMMARY OF N OTABLE LTE E N ODE B S OFTWARE I MPLEMENTATIONS

that reconfiguration of hybrid FPGA is a very slow process and The recently approved 3GPP non-standalone 5G standard20
it is not straightforward to do it on the fly. talks about the scenarios where LTE’s eNB and 5G NR’s (New
1) Lessons Learned: A summary of notable software Radio) gNB (next generation NB) co-exist with EPC and/or
implementation of LTE eNodeB is provided in Table II, where NGC (Next Generation Core). The coexistence is possible
it is very clear to see the current state of the art. All of the LTE through evolved base-station eLTE eNB which can connect to
eNodeB software implementations, open and closed, are rela- EPC as well as NGC and new interfaces, such as, Xx (between
tively recent and unless independent and comprehensive trials LTE eNB, connected to EPC, and gNB) or Xn (between
are done, it would not be clear that they have resolved the eLTE eNB and gNB with either one or both are connected
latency issue of GPP or not. Although, they definitely show to NGC) [57]. NGC is a step towards a fully softwarized
a lot of promise and potential of the technology. We expect core by complete separation of control and user plane and
that like every technology, GPPs will also improve and multi- modularizing the management functions. The non-standalone
core processors will resolve the bottlenecks slowing down their 5G allows an intermediate phase with backward compatibility
speed. to LTE.
Moreover, as we have observed while discussing fronthaul, Hybrid SDN network where traditional or legacy nodes
that leaving PHY at the RRUs could save the capacity of co-exist with SDN-enabled nodes have been closely investi-
fronthaul and it is very difficult to imagine that network oper- gated in case of wired and fixed networks and the state of
ators would open such low-layer networking to the third party the art is way ahead of respective mobile network research.
users. We really do not loose much if PHY, which is also very Firstly, an important point to note here is that OpenFlow
computationally intensive and latency critical specially in 5G, protocol, which enables the switches to act as dump data
remains in special hardware at the base-stations or RRUs, at pipes by following a flow table provided by the controller,
least in the initial phase of SDN roll-out. can also be implemented on commercial Ethernet switches
and routers [58] making them SDN-enabled. The simplest
approach could be the dual stack approach where all network
nodes have SDN interface along with an interface for normal
C. Backward Compatibility processing [58]. This approach, however, requires substantial
Not long ago, the carrier networks have deployed their 4G effort to implement the software on each and every switch of
LTE systems. The systems are performing very well and get- the network.
ting the most out of them by enhancing their capacity and Moreover, there are a number of proposals which has shown
support for number of users is a straightforward business different mechanisms to successfully operate hybrid SDN
opportunity. SDN technology comes with big promises but networks with some SDN/OpenFlow switches and many non-
we cannot expect that the carrier operators and owners will SDN or legacy switches [59]–[62]. The performance of hybrid
decommission the recently-erected network. Most probably, SDN networks show real promise and the architecture can
4G-LTE equipment will remain in function for some num- actually provide a long-term solution rather than a quick fix
ber of coming years. In view of this, any new technology for the transitional period. In [59] a small number of SDN
should have the essential feature of co-existing and working switches, strategically placed, are shown to be sufficient for
with the legacy systems. In this section, we will review the the network to reap the benefits of SDN and behave like a full
work which has been done on the compatibility issues of SDN- SDN network by enforcing SDN policies, e.g., access control.
based mobile networks and legacy systems. In legacy systems,
co-existence with 4G equipment is the most important. 20 http://www.3gpp.org/news-events/3gpp-news/1929-nsa_nr_5g
ZAIDI et al.: WILL SDN BE PART OF 5G? 3231

In [61] and [62], the legacy switches are manipulated by spe- base-station has a proxy serving as GTP (Gateway Tunnel
cial messages to forward the packet over the desired route. Protocol) end-point. Softcell implements LTE core network
10% penetration of SDN switches are shown to be sufficient using commodity switches which then carry normal IP traffic
to implement failure detection and route recovery mechanism between base-stations and Internet. Softcell, however, does not
through SDN controller in [60]. These approaches for manag- talk about softwarization of the base-station.
ing hybrid networks, however, incur significant management Heterogeneous CRAN [33] also claims backward compat-
complexity, as they control legacy and SDN switches via dif- ibility with legacy cellular systems and other CloudRANs as
ferent mechanisms [61]. Moreover, the existing techniques on along with RRUs and centralized BBU pool, it also introduces
hybrid SDN networks successfully show implementation of some high power nodes or base-stations providing control sig-
SDN policies of routing, access control, flow isolation, failure naling. These high power base-stations are similar to macro
recovery, etc., but it is not possible to efficiently virtualize the base-stations. According to Peng et al. [33], these nodes are
legacy resources via SDN controller without using additional critical to guarantee backward compatibility where multiple
virtualization mechanisms such as hypervisors etc. heterogeneous radio networks converge and this architec-
All the above discussed schemes are designed for fixed ture can take advantages of both CloudRAN and cellular
networks and not for radio access network where the SDN- networks.
based RRUs will need to co-exist with eNodeBs, 2/3G Similar ideas were presented in [67] under the title of
base-stations, and RRUs of CloudRAN. RRU of CloudRAN Fusion Net, where macro cells comprise of host layer pro-
may not exactly be the same as RRU of SDN-based mobile viding anchoring points to the users. However, the traffic is
network. In the context of mobile networks, the research dynamically transmitted and aggregated from a boosting layer
is way behind than the fixed wired networks and there are consisting of Wireless LAN (WLAN) access points, small
only very few attempts to propose solutions for the backward cells, RRUs, etc. The paper also discussed SDN and NFV as
compatibility issue. emerging technologies but did not bring them in the context
Mobileflow [63] is one of the earliest solutions to resolve the of Fusion Net.
compatibility issue between SDN-based networks and legacy In [68], the discussion about backward compatibility leads
systems using virtualization techniques. It defines a mobile to the mechanisms where SDN-based network (RAN and core)
flow controller which can be used to implement the LTE can be configured to be used as LTE 4G network. This is ben-
control plane via network applications. The network appli- eficial if 4G UE (User’s Equipment) is being served by the
cations provide all LTE control plane functionalities, e.g., network. However, the UE should have the appropriate appli-
control for S-GW (Serving Gateway) and P-GW (PDN or cations to instantiate control and data transmissions. Moreover,
Packet data network Gateway) etc. Forwarding plane is defined the system still does not show the co-existence capability with
through mobileflow forwarding engine which work similar to 4G system where seamless handover and resource manage-
OpenFlow switches by following the routing rules given by ment could be possible between the two systems. According
the mobileflow controller. They also have radio interfaces to to [69], the main property for a SDN enabled backward com-
connect to UEs. The authors claim that operators can use patible network is to implement standardized interfaces for
this architecture in one part of the network to be used with interworking with legacy networks.
the legacy equipment and then move to a more flat con- Another effort of addressing backward compatibility issue
trol model, without LTE control plane entities, in future by is presented in [70] where a 4G mobile communication system
changing the virtual machines. They have validated the archi- is shown to be used as a SDN-based mobile network via the
tecture through a prototype. However, the most important part separation of Edge Control Plane (ECP) and edge switches
of mobile network, i.e., eNodeB is not decomposed into the from the 4G core network. ECP contains all the intelligence
control and data plane and it is not shown how they will be and LTE control plane, i.e., MME, HSS, PCRF, etc., where as
mapped into the mobileflow forwarding engine and associated the 4G network only performs packet transmission. The access
radio interface. In the prototype, an eNodeB was remained nodes, i.e., eNodeB, remain unchanged. Reference [70] also
as it is and the mobileflow forwarding engine was used as talks about serving 3G and WiFi nodes through similar SDN-
the first node after eNodeB to direct the traffic. Although, not fixed 4G network. This scheme is a good proposal for first
explicitly mentioned, mobileflow is in fact a SDN-based core step towards the full SDN implementation but it requires sub-
network alternative to be used in place of 3G/4G core networks stantial management rework in 4G networks. It is yet to see if
with option of moving towards a full SDN network in future it will make a good business case to revamp the control plane
through software update on the controller. The details about of already erected 4G network or will it be easier to install
how to implement EPC through OpenFlow switches are also the new 5G systems, already arriving in the market, with sim-
mentioned in [64]. ilar SDN fixes as in [70]. The proposed SDN-based mobile
In optical fiber domain, [65] showed SDN-controlled opti- network architecture in [71] maintains the 3GPP interfaces
cal topology-reconfigurable mobile fronthaul achieving 10Gb/s so the architecture can also be used with legacy network.
peak rates with <7µs back-to-back transmission latency and The architecture moves EPC to cloud but leaves eNodeB
29.6dB total power budget while maintaining backward com- unchanged.
patibility with legacy fiber deployment. 1) Lessons Learned: In order to summarize the issue and
A proxy based solution for backward compatibility of state of the art of the relevant solutions, we believe that
SDN-based core network is proposed in Softcell [66]. Each interoperability or backward compatibility may take different
3232 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

definitions but coexistence with 4G or legacy networks should For the radio access networks, schemes discussed in the
include the following scenarios: previous section and introduced in [68] and [70] provide the
1) A 4G or legacy UE in the coverage of SDN-based initial approaches towards evolutionary deployment of SDN-
mobile network. Reference [68] tried to look at this type based mobile network building operators’ confidence before
of scenarios where a SDN-based network can be pro- major upgrade of the system. SDCN [9] talks about a phased
grammed to serve a legacy/4G UE, although, the UE is approach towards fully programmable centralized RAN. In the
required to have appropriate application. This scenario first phase, the control or signaling plane of legacy systems,
is, however, not really interesting from practical point i.e., radio resource scheduling, hand-off, paging, etc., is moved
of view since the upgrades in mobile handsets are much to a logically centralized location or cloud. In the second phase
more rapid than the enhancements in the access network. the data planes of the base-stations can be implemented by
2) A legacy base-station or eNodeB is attached to a SDN- SDR (Software Defined Radio) on commodity of the shelf
based core network. As shown in [68], the SDN-based (COTS) hardware. In the third phase CloudRAN is imple-
network can be programmed to serve as a 4G network. mented with BBU pool using general purpose platforms and
In Softcell [66], the base-stations are assigned proxies RRUs on the remote site.
which can act as GTP end-points to communicate with Similar step-wise idea is presented in [71]. The proposed
SDN-based core network. The major property identi- SDN-based mobile network architecture in [71] suggests a 3
fied in the literature for interworking of new and 4G step migration process from legacy network to fully complaint
systems is to maintain the standard 4G interfaces and SDN network. It moves EPC to cloud while maintaining 3GPP
GTP tunneling [69]. In 3GPP’s 5G non-standalone archi- interfaces with legacy nodes. In the second step, OpenFlow
tecture, this scenario is similar to option 4/4a/7/7a [57]. switch is introduced and GTP tunneling is replaced with
The solution is to enhance eNB and define additional MPLS tagging. In the third step, legacy nodes (S/P-GW) are
interfaces. The standard also defines the situation when removed from the network. The eNodeB stays the same in this
a 5G gNB (next generation NodeB) could be connected architecture [64].
to an EPC through assistance from LTE eNB using a Softcell [66] proposed a SDN based LTE core network using
new interface Xx. commodity switches and routers. The major contribution lies
3) eNodeB (eNB) or heterogeneous legacy base-stations or in compressing thousands of rules and policies into manage-
phantom cells and RRUs of SDN-based mobile network able routing rules for off-the-shelf switches. They proposed
share a geographical area. The SDN network should take that the carrier put a proxy on each base-station which can
the spectrum resources used by the eNodeB or legacy act as GTP end-point. The Softcell switches then carry nor-
cells into account while distributing resources among its mal IP traffic between base-stations and core network. The
RRUs. According to our search, no such scheme exists network can have Softcells in one part while legacy core in the
in the literature. rest in the initial phase and can slowly move to full Softcell
4) Handover between 4G and SDN-based mobile network. implementation. The proposal, however, does not cover the
Mobility management mechanisms should be aware of radio side of the network and base-stations are not made
the possible handover between the systems. For SDN- programmable.
based core network, Softcell [66] suggests implementa- Mobileflow [63] also allows part of the core network to
tion of S1-MME and S10 interfaces so it can handover be SDN-enabled but still work with legacy networks and
to and from legacy LTE EPC. 3GPP’s recently approved base-stations. The architecture supports GTP tunneling and
standard for non-standalone 5G NR (New Radio) defines maintains the EPC standardized interfaces for compatibility.
the handover between multiple RAT (Radio Access Independent implementation studies are required to confirm
Technologies). The solution assumed a new interface the claims of Softcell [66] and Mobileflow [63] that they can
between EPC and NGC (Next Generation Core) [57]. co-exist and work with legacy systems. Moreover, it is impor-
tant to check the performance of such hybrid systems and how
easy or difficult it is to implement SDN rules and policies on
the flows.
D. Disruptive Deployment An interesting classification of proposed SDN-based mobile
The issues of disruptive deployment and backward compat- network architectures based on revolutionary or evolutionary
ibility have a lot in common and the schemes discussed in the migration is done in [72]. Most of the evolutionary architec-
above section are in fact making it possible to deploy SDN tures are shown to have preserved GTP tunneling. OpenFlow,
networks in an evolutionary fashion. A good example is the in general, does not support GTP tunneling and that is quoted
recently approved 5G NR non-standalone standard which is as a major reason to use MPLS labeling [71] in SDN networks.
an intermediate step towards full 5G NR deployment and it Although, there are attempts to extend OpenFlow to support
allows the operators to start experimenting with 5G equipment GTP tunneling [64]. Reference [69] generalizes the evolution-
right away. However, the work on fixed networks are much ary approach by identifying the importance of maintaining
more mature and sophisticated than wireless access networks standardized interfaces by SDN network to have interwork-
and there are a number of investigations which showed that a ing with the legacy networks. Similar considerations were part
small number of SDN switches are sufficient for the network of the evolutionary approach in [73] to move towards a flex-
to reap benefits of SDN in the transition phase [59]–[62]. ible architecture for RANaaS (RAN as a Service). RANaaS
ZAIDI et al.: WILL SDN BE PART OF 5G? 3233

chooses level of centralization according to the requirements controller, forwarding plane and the links between them are
and network state. exposed to new threats and attacks.
Three different alternative are being discussed for SDN Another important issue with SDN networks is the config-
deployment: overlay, white-box, or custom hardware.21 uration errors [80]. The network allows users and application
Overlay technologies on existing infrastructure, such as, hyper- to configure and program the network which may result in
visor for virtualization will offer minimal disruption but will inconsistent policies and flow rules which could be both inten-
not provide flexibility to use the hardware efficiently. Some tional or unintentional. It is very important to constantly check
optimization opportunities may be missed. Vendor-specific the network for inconsistent policies without adding a lot of
customized solutions using ASIC can also be deployed with monitoring overhead and overloading the controller with mes-
minimal interruptions but will not provide a truly real-time sages. Scalability and availability of the controller is also an
programmable and economical platform. On the other hand, issue with logically centralized SDN control plane. This issue
white-box solution requires a lot of standardization across the is also discussed and looked at in the security related litera-
hardware and software domains to simplify the deployment ture. A simple solution is to have distributed control where
and systematize the upgrade procedures. each controller is responsible for a cluster of switches with
1) Lessons Learned: Most of the ideas of evolutionary some overlap with other controllers to have robustness and
deployment of SDN-based mobile networks remains at a high- resilience against failures [80]. Moreover, SDN are developed
level and prototype implementation are necessary to ascertain over GPP platforms which may have their own vulnerabili-
their workability. From business perspective, the point solu- ties. It is important to harden the platform or it will become
tions which can work with minimal disruption but can result potential surface for attackers [81].
in quick monetization are the most favorable. Now, the ques- Similar to other aspects of SDN, the work on security of
tion is: what these point solutions will be in the SDN context SDN-based wired networks are more mature than in wireless
filling in the puzzle pieces to complete the bigger picture. SDN networks [74]. Although, the work is equally relevant for
The idea of non-standalone 5G NR (New Radio) in 3GPP’s SDN-based mobile networks as even in the wireless settings,
recently approved standard [57] is an intermediate step and the SDN controller and links connecting it to the switches
a good trend to follow. The transitional phase is achieved are still considered to be the most vulnerable points for
through additional interface designs and evolving current LTE attack [74], [82]. The wireless part of the system, however, has
eNB into eLTE eNB which can connect to 5G core. 5G NR some unique issues, such as, user mobility, the co-existence
is a step towards full softwarization of the network through of various generations of cellular systems and WiFi owned
separation of control and user plane and modularizing of by different operators, security of wireless link over shared
management functions. medium, etc., [77].
Surveys of SDN-based networks’ vulnerabilities and related
solutions are presented in [74], [75], [79], [80], and [83].
E. SDN Specific Security Issues Survey in [74] identifies possibilities of security threats and
The openness promised by SDN also introduces additional attacks in SDN in relevance with the traditional STRIDE
security risks and vulnerabilities [1]. Currently, carriers do (Spoofing, Tampering, Repudiation, Information disclosure,
not allow third party to meddle with the network protocols, DoS (Denial of Service), Elevation of privilege) threat
routers, and switches, which results in an ossified network but model and discusses respective work and proposed solu-
its management and security is comparatively easier. Users’ tions. According to [74], DoS (Denial of Service) attacks are
access to the networking fabric will increase occurrences expected to be frequent in SDN networks than in the traditional
of intentional and unintentional incidents affecting network networks as significant amount of data will be exchanged
operations. Moreover, the logically centralized controller is between the central controller and the switches. Spoofing may
very critical component of the network. If the controller is have less chances of occurrence in dynamic SDN network
compromised, the attacker will have tremendous access over as it depends on tricking a network services based on obso-
the network [74], [75]. On the other hand, data is not co- lete information [74]. We have looked at each SDN layer and
located with control plane as in traditional network and the interfaces and discussed only the important issues identified
attacker needs to re-direct the data to get hold of it [74]. in the literature. A comprehensive survey of security related
Moreover, the SDN virtualization abstraction layer integrates work is outside the scope of present paper and we refer the
various ISP platforms while hiding the specific protocol details readers to [74] and [80] (more recent) for thorough survey
which makes security management comprehensive and man- of security related research in SDN networks. A comprehen-
ageable [76]. Furthermore, the capability of redirection or sive discussion about DDoS attack and counter-measures is
filtering of the flows based on packet content comes natu- available in [79] for SDN networks. The survey in [83] is
rally with SDN and can enhance the network security [77]. exclusively for mobile networks.
According to [78] and [79], global network view, self-healing 1) Controller: The logically centralized controller is very
mechanisms, and the additional control capabilities give SDN critical component of the network. If the controller is com-
great security advantage over traditional networks, though, the promised, the attacker will have tremendous access over the
network [74], [75], [77], [83]. The idea of standby controller,
21 http://www.enterprisenetworkingplanet.com/netsysm/making-the- when one of the controller fails, is discussed in [76]. There are
business-case-for-sdn-1.html several proposals for distributed control design for scalability
3234 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

and reliability [80], although, the distributed control plane and References [81] and [83] point out that the packet without any
its interface for interworking are not properly explored. A match to the flow table is sent to the controller and an attacker
Byzantine mechanism is also proposed where each network can exploit this vulnerability and overwhelm the controller just
element is managed by multiple controllers. Optimization is by sending data. A possible solution is to design switches with
used for controller assignment to minimize the number of enough processing power to forward maximum expected load
controllers for a given Byzantine fault tolerance [80]. and not plenty of packets [81]. According to [85], SDN devel-
A simple DoS or DDoS (Distributed Denial of Service) opment still need to go a long way to achieve the required
attack can exhaust resources in the controller and also in the security and dependability.
forwarding switches. Authentication for legitimate users can 2) Southbound Interface: Dedicated special connection can
mitigate this issue up to an extent but a compromised applica- be provided to improve security of the links between con-
tion with legal credentials can inject faulty flows causing DoS. troller and its switches, also called the southbound interface,
A flow-based anomaly detection mechanism can resolve this but it does not eliminate the possibility of compromising the
issue [74], [83]. Reference [84] develops a connection migra- communication [74]. The term southbound may suggest direc-
tion technique on the data plane to classify successful TCP tion of flow from the controller to the switches, although, this
connections from unsuccessful ones in order to protect con- interface is bidirectional and we are using the terminology to
trol plane from saturation attacks. Data plane can send network conform with the SDN research community.
information to the control plane asynchronously via actuat- Most of the solutions developed for SDN are designed
ing triggers. According to [80], with SDN characteristics of specifically for OpenFlow [74], although, the threat vectors
dynamic flow table management and distributed control, the are independent of any technology and based on the con-
threat of SDN-specific DoS attack can be reduced. ceptual layers of basic SDN architecture [85]. An important
On the other hand, the SDN controller provides unprece- issue with OpenFlow is that Transport Layer Security/Secure
dented opportunities to develop adaptive and dynamic counter- Socket Layer (TLS/SSL) encryption is optional and not
measures for security threats. Although, the dynamic security enforced [74]. Moreover, TLS/SSL based communication is
control which can adapt to the changing networking condi- not strong enough to protect the control channel from IP based
tions may also pose new challenges unknown to conventional attacks [82]. Furthermore, the digital SSL certificate used for
networks [74]. A survey of possible ways of launching DDoS the communication is self-signed and is not very secure [81].
attack and the solutions are presented in [79]. The major If the private key of the certificate is stolen, the attacker can
emphasis is on the capabilities of SDN, such as, global infor- eavesdrop and join the network.
mation, dynamic update, centralized control, and software- If SSL encryption is not used, it is possible to launch an
based traffic analysis, to mitigate the attack. The survey ARP (Address Resolution Protocol) spoofing attack on south-
identifies open questions in this respect. bound interface where an attacker’s MAC address is linked
A lot of work has been done in SDN-based firewalls which to a legitimate IP address. Counter-measures to ARP spoof-
are mostly implemented as northbound API to the controller. ing attack include packet-level monitoring [74]. Similarly,
The controller stores and dynamically update the Access lack of encryption can also cause Man-in-the-Middle (MiM)
Control List (ACL) whereas the traditional firewalls are static attack. MiM can misdirect and drop flows and can scan the
and may contain obsolete rules as they are manually entered network and compromise confidentiality. Alternative encryp-
by network administrator [74]. In the existing security model, tion methods along with proper auditing and logging methods
policies are in place to enforce the traffic to physically go are suggested to counter MiM and verify non-repudiation [74].
through security middle-boxes and firewalls. In SDN, how- In a distributed controller environment, where each con-
ever, the paths are logically chosen by controller and it is the troller is responsible for a cluster of network switches, the
responsibility of the controller to ensure that firewalls etc., are interface between controllers to share important information
enforced [81]. is also vulnerable similar to the southbound interface and can
Reference [80] surveys the access control mechanisms for have similar types of attacks [74], [83]. As far as OpenFlow is
OpenFlow where permission systems are developed to apply concerned, this interface is not properly researched and stan-
minimum privilege to the application instead of the cur- dardized yet and it is still to see what shape the interface will
rent state where full privileges are open to all applications. take.
Similarly, methods around enforcing the permissions are devel- 3) Northbound Interface and Application Layer: The north-
oped to secure the interface between controller and application bound interface is also exposed to significant threats as the
along with authentication methods to stop unauthenticated user applications with power to access and define network
users to access the controller. A summary of possible security resources and processes can also abuse the network intention-
threats, causes and solutions is presented in Table III. ally or unintentionally [74]. There is a lack of standardization
Another issue raised in [85] is about the resilience problem for application access management regarding rules for grant-
of the current controller implementations, such as, Floodlight, ing privileges, binding mechanisms, auditing, trust mechanism
OpenDayLight, POX, and Beacon. It is claimed that com- between network management applications and controller,
mon application bugs are enough to crash existing controllers. etc., [75], [77], [83].
From the security point of view, a simple malicious action, In order to prevent any malicious application to gain access
such as, changing the value of a data structure in memory can to the controller and network, some advanced solutions are
directly affect the working and reliability of the controller [85]. also proposed besides establishing trust and authenticating the
ZAIDI et al.: WILL SDN BE PART OF 5G? 3235

TABLE III
S UMMARY OF I DENTIFIED V ULNERABILITIES AND S OLUTIONS

user [80]. These solutions uses techniques, such as, resolv- number of unknown packets will exhaust the resources at
ing the conflicting rules based on the priority of application, the switch.
securing the controller core by sandboxing each application, In OpenFlow, VAVE (Virtual Address Validation Edge)
and creating an isolation layer between application and control module verifies the address of external packets with no record
plane for fault isolation [80]. in the flow table. The module is extended to include binding
4) Data Plane: Forwarding switches are vulnerable to dif- validation and a data-set of black and white lists to counter
ferent types of attacks, such as, DoS, modification or injection IP spoofing attacks [74]. Random and continuous change of
of fraudulent flow rules, compromise, etc. A compromised host identities are also proposed as a counter-measure where
switch can discard, slow down, or deviate network traffic [77]. an attackers scans the network to learn flow patterns [74].
Even worse, it can launch a DoS attack on the controller [77]. According to [85], some attacks, such as spoofing, is not spe-
The survey in [83] also mentions the possibility of flow table cific to SDN but has a much larger impact in the SDN. The
overflow in the switches due to an intelligent attacker sending authors give an example of spoofing the address of the con-
flow requests with slight difference in parameters. troller and comment that a smart attack which would last only
Yan et al. [79] discuss DDoS attack possibility on all for few seconds just enough to install malicious flow rules is
SDN layers including the controller. For example, a new and very detrimental and very hard to detect.
unknown flow to a switch will be referred to the controller and The vulnerabilities of a special type of SDN network,
even if only the header information is sent to the controller, called stateful data plane SDN, where some programmability
the packet will be saved at the switch. An overwhelming is offloaded to switches, such as, rules to change forwarding
3236 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

tables, are discussed in [86]. The additional functionality discussed before, most of the work is mainly focused over
improves the latency of the network and reduces the load at OpenFlow and its mechanisms. Table III is definitely not
the controller but also arises the issue of asynchronous and complete and as the state of the art moves forward, new
inconsistent states at the switches. Frequent updates to the threats and vulnerabilities will be identified. Reference [75]
controller can also be exploited to launch a saturation attack. presented a similar table but we have added information from
In a virtualized environment, it is also very important to other publications into the table and talked about the possible
build tools which can verify logical separation between vir- solutions to have clear idea about the current state of the
tual networks. In practical networks, a VM (Virtual Machine) art. Some attacks can be classified differently than Table III
may access some resources of another VM sharing the same considering the multiple layers and interfaces it may affect
physical platform, tampering the network information [74]. as also noted in [80].
Formal verification based methods are being proposed to ver- 5) Issues Exclusive to SDN-Based Mobile Networks: The
ify the isolation of traffic between network slices [80]. So far, security analysis and counter-measures discussed above are
the research is focused on OpenFlow and should be extended designed for wired networks only and mobile environment
for broader SDN deployments. features and cross-domain interoperability is not considered
Configuration issues where intentional or unintentional pro- in any of them [76]. In this section, we highlight the secu-
gramming of the network causes conflict in policies and flow rity issues and solutions specifically for cellular and wireless
rules are also termed as a major problem specific to pro- networks. As can be seen in this section, security of mobile
grammable SDN networks. Not just applications, but multiple networks has not gained a lot of attention in the research
entities with write access can override or misconfigure the community.
related flows and compromise the switch functionality [80]. a) Vulnerabilities: References [77] and [76] identify
As new applications and new devices are added into the four major issues in SDN-based mobile networks besides the
network, security techniques should be in place to make sure threats and attacks identified for the wired part of the network.
network correctness and operability [80]. According to [80], These issues are:1) mobility between different access technolo-
the area of configuration issues is one of the widely researched gies using different security protocols, 2) mobility between
theme in SDN and solutions are proposed for detection of networks owned by different operators, 3) highly constrained
network errors and data plane verification. These proposals computational capacity and storage, and 4) backward compat-
can be classified as real-time or non-real-time. In non-real- ibility to older generations of mobile systems. These issues
time solutions, symbolic executions are used to test OpenFlow create unique security problems in mobile network domain.
applications for correctness and not causing the network to The survey in [83] looked exclusively at the security issues
reach an inconsistent state. Binary decision diagrams are used of SDN-based mobile networks. The paper classified all
to check intra-switch misconfiguration [80]. Flow policies and attacks in terms of SDN layers as also done in the present
flow requests are also verified not to by-pass firewalls. These paper and extends the scope to mobile terminal and radio
verifications may take several minutes and are not designed for access medium. Mobile terminals are in ever increasing danger
real-time anomaly detection [80]. There are real-time tools as of Trojans and viruses. The terminals are resource constrained
well for verification of flow tables (e.g., identifying routing in terms of computability and storage and cannot execute
loops, unavailable paths, etc.), conflict resolution for applica- heavy duty intrusion detection modules [83]. It is important
tions in a firewall, and detection of firewall violations [80]. to customize the traditional tools to suit mobile terminals’
According to [80], the real-time tools show the evolution in need of light weight safety procedures. Malicious or legiti-
right direction, although, experimental deployments will be mate RF interference can cause saturation attack at the access
needed to bridge the gap between theoretical models and points. Threat of DoS from base-stations to the core network
characteristics of live network. is considered low in [87] as the traffic from the users to the
The survey [80] also lists the northbound APIs which are network or the base-station control traffic is generally low.
designed for conflict resolution in policies. They extract non- Reference [83] also surveyed the security literature in SDR
overlapping policies from the flow rules before instructing the (Software Defined Radio) area and identified RF interference,
controller to install the flow rules into the switches. The work, mobile terminal malware, and MAC tempering as some of the
however, should be extended for distributed control environ- detrimental attacks in wireless environment.
ment. Policy management techniques are also proposed for Mobile user’s privacy and confidential information can
conflict resolutions across multiple modules in different lay- be leaked in an open backhaul network architecture [87].
ers and also for comprehensive policy update across all the Currently, the traffic is tunneled using IPSec to the core
relevant packets and flows [80]. According to [80], a con- network to secure the mobile terminal’s identity from any
cern with these policy conflict resolution is their scalability for eavesdropper. Similarly, important control traffic can also be
larger applications or larger networks as all of these techniques compromised over open backhaul. Another issue raised in [87]
are highly computationally extensive. Another alternative is to is about a malicious access point or base-station gaining access
maintain a strongly consistent data store, instead of conflict to the network by spoofing its location using a legitimate base-
resolution in policies, to maintain network consistency. station location. Counter-measures based on GPS tracking can
A summary of identified attacks and their solutions with be developed to mitigate the attack. Moreover, the small base-
relevance to SDN-based networks and specially SDN-based station/access point/remote radio unit is also vulnerable to
mobile networks is presented in Table III. As we have physical tempering.
ZAIDI et al.: WILL SDN BE PART OF 5G? 3237

b) Secure architectures: Some work has been done in A framework for security and trust for 5G is proposed
the SDN-based mobile network area which is also focused in [92] in the context of virtualization and SDN. The
on OpenFlow. Reference [88] adds vertical forwarding exten- authors argue that security mechanism in cloud computing
sion to OpenFlow to deal with mobility and access control are not sufficient for 5G security challenges and required
management while forwarding to legacy elements of network. extensions to provide sufficient security services and func-
Reference [89] identifies the lack of secure mobility support tions based on NFVI (NFV infrastructure) in a trustworthy
which make OpenFlow inapplicable in wireless and mobile and economical way with cooperation of various trustwor-
systems. OpenFlow cannot handle switch mobility, e.g., in a thy VNFs (Virtual Network Functions). A security and trust
moving train, and secure change of IP addresses in a fast and framework is proposed within NFV MANO (Management
dynamic mobile network to maintain secure communication and Orchestration) architecture with NFVI-TP (NFVI Trust
between the controller and the switch. The solution in [89] Platform) embedded in NFVI by an authorized party. It ensures
is based on HIP (Host Identity Protocol) security method to NFVI platform layer security by providing a root trusted mod-
identify a host either by a host identifier or a host identity tag. ule, a hardware resource, to ensure every component build
A flow control agent is also used in [89] to update controller upon it is certified as trusted. The framework is not evaluated
of the new location information for location based services. through implementation though. Moreover, the framework and
Reference [82] proposed an architecture for SDN-based associated discussion is generic enough for fixed networks
mobile networks using HIP (Host Identity Protocol), IPSec and does not address any particular issue regarding mobile
tunneling, and SecGW to provide secured control channel. networking.
Although, the evaluation used OpenFlow but the proposed The idea of taking benefit of SDN features to provide secu-
architecture is independent of SDN protocol. Resilience rity services is used in [93], where a low-latency and simple
against various attacks, such as, DoS, IP spoofing, eaves- authentication handover scheme is proposed for 5G dense
dropping, etc., is analyzed. The architecture is shown to be HetNets. The major benefit comes through the centralized con-
vulnerable to volume based DoS attacks and an access con- trol of SDN which is also assumed to host an authentication
trol method or firewall is recommended to be used to prevent handover module. Both access points and user need authen-
such attacks. In a follow-up work a software defined moni- tication before accessing the network. The simulation results
toring and data collection scheme is implemented on top of show low handover delays than the traditional method.
the secure control channel to prevent, detect, and react to the In a recent publication [94], an SDN based 5G archi-
security attacks [75]. tecture is proposed with three different OpenFlow gate-
Reference [76] proposed a SDN-enabled security architec- ways and controllers, one each for the mobile user, for the
ture for mobile networks where a security layer is introduced RAN, and for the network. The network is comprised of
above the control layer to monitor the application and stop heterogeneous technologies. The authors identified potential
the malicious traffic from entering the radio environment. They threats and proposed security mechanisms, i.e., authentica-
proposed local agents at the wireless routers and access points tion, access control, periodic key updates, routing conflict
for fast responsiveness. Since SDN hides the complexities analyzer, for communications between all types of the archi-
of multiple domains and their specific protocols, it is rela- tectural entities and showed the proof of their feasibility and
tively easier to provide comprehensive security through SDN strength.
architecture. A summary of the secure architectures discussed above
Reference [90] proposed a three-level control plane struc- is presented in Table IV, where the major security feature
ture for SDN-based mobile networks to implement a unified (or features) and important architectural attributes are also
security, connection, mobility, and routing mechanism. They highlighted. As it be seen from the table that access con-
have included device controller (for UE), an edge controller, trol is one of the popular mechanisms to secure northbound
implementing 5G network functions through control applica- interface. Also, more detailed solutions emerged as extensions
tions, and an orchestration controller to coordinate utilization to OpenFlow and the once claiming to be independent of
of cloud resources. Two versions of edge controller are iden- the physical infrastructure have used OpenFlow for proof of
tified in the paper, one implemented in the cloud and the concept. The current state of the art with respect to secure
other in the mobile device for mission-critical control and SDN-based mobile networks still needs solutions for broader
out-of-coverage services. Without going into the details of the spectrum of threat models including all types of DoS attacks
network functions, the paper describes that security along with along with comprehensive evaluation of robustness of the
radio resource management, mobility, etc., is implemented solutions.
though control applications by the edge controller. c) Tools: Reference [95] proposed a security assessment
Reference [91] addressed the computational limitations of scheme to quantify security levels to networks. The scheme
small wireless devices by proposing a light-weight security is proposed for mobile SDN networks although it is generic
framework for D2D communication. They define spatial trans- enough to be used in other networks. The security level of a
mission region for D2D link that can guarantee a minimum network is defined in terms of efforts to reach a target error
secrecy rate through exploiting the physical characteristics of state or in launching an attack. More efforts required to launch
the wireless channels. Analysis shows that as number of users an attack would represent resilience or high security level of
increases, the achievable transmission capacity also increases the system. The authors want to explore alternative methods
due to availability of more D2D pairs. for quantification of effort in their future work.
3238 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

TABLE IV
S UMMARY OF S ECURE SDN-BASED M OBILE N ETWORK A RCHITECTURES

Security issues in embedded mobile devices are discussed control plane is an important paradigm for reliable and
in [96] and a learning intrusion detection system is proposed scalable network control. This paradigm is not explored
for OpenFlow networks. Statistically different traffic than in detail and the inter-controller interface is not yet
the user-defined normal traffic is termed as anomalous. This defined.
work is not directly related to cellular systems rather it is 3) Access control There is still lack of standardization
designed for embedded mobile devices with strict constraints for applications to access the controller. The permis-
on transmission power and computational capacity. The idea sion schemes [80] should be thoroughly evaluated for
of intrusion detection system to detect DoS though has appli- correctness and computational feasibility.
cation in cellular scenarios. A formal verification method is 4) Southbound interface A dedicated link can connect
developed for OpenFlow to make sure systems’ correctness controller to its switch but appropriate level of encryp-
and security [97]. Malware protection for mobile devices based tion is needed to provide secure communication. The
on OpenFlow protocol is discussed in [98]. The protection monetary implications of dedicated link and feasibility
system uses real traffic analysis of connection establishment should be studied.
packets inside an OpenFlow controller. 5) Isolation between virtual networks Real-time tools are
Considering the storage and computational limitations of needed to verify complete logical isolation between the
mobile devices for very heavy security applications, [99] virtual networks sharing the same physical resources.
proposed outsourcing security to cloud using OpenFlow virtual 6) Error/anomaly detection Anomaly or error detection
switch in the device and OpenFlow controller residing in the is an important element in the network where there is
cloud along with security manager. This method takes away high risk of controller/switch compromise. The tech-
major computational load from the device, employs variety niques should be tested for overhead, detection accuracy
of security services with fast processing. The communication and time, and it is also necessary to secure the anomaly
delay is an important factor affecting performance and so far detector from security attacks.
it is not clear if faster processing overcomes the transmission 7) Policy conflict resolution The SDN network allows
latency or not [99]. users and application to configure and program the
6) Summary of Outstanding Security Issues: A compre- network which may result in inconsistent policies and
hensive survey of SDN security solutions have been done flow rules. There are real-time tools for policy control
in [74] and [80]. However, the solutions are still not mature resolution but they have scalability issues while dealing
enough to be used for production deployment. Here we list with large applications or large networks.
some of the take-away points from our survey of security 8) Data leakage and modification It is shown in [80],
literature for SDN networks. that there are no solution till date for data leakage and
1) Comprehensive evaluation Although, a lot of work modification is SDN networks. However, encryption is
has been done in SDN security but independent exper- sought to be a valid solution for SDN networks as in
imental evaluations and comprehensive assessment of traditional networks but the physical separation between
solutions are required before the technology is ready for controller and switches requires very important control
deployment. information to be pass over the communication links.
2) Controller security Controller compromise and mal- Their security should be comprehensively tested for the
function can cause huge damage to the network. DoS available encryption standards.
or DDoS attacks on the controller are the simplest and 9) Solutions conforming to SDN principles SDN
most expected attacks in SDN environment. Distributed networks are supposed to have multi-vendor
ZAIDI et al.: WILL SDN BE PART OF 5G? 3239

interoperability, third party application support, to [100], TCO (Total Cost of Ownership) of traditional virtual-
integration of virtualization [80], and fast and dynamic ization approaches tracks the growth in requirements whereas a
reconfigurations. Any security solution proposed for common standardized platform based virtualization, e.g., inte-
SDN should also conform to the norms of the network. grated SDN and NFV, breaks the linkage between cost and
requirement and can have 62% lower TCO than traditional
F. Clear and Compelling Business Case approaches.
A ROI analysis by ACG Research [100] claims that the
Despite the tremendous work in SDN and towards its
break-even point can be achieved in one year when traditional
deployment, an operator survey conducted in 2015 reveals that
virtualization solutions are gradually phased out in favor of
the majority thinks that lack of clear compelling business cases
a common and standardized platforms, e.g., SDN, for NFV.
for SDN is the major obstacle in its widespread adoption and
This analysis considers network transformation cost including
roll-out. The deployment cost came out as the second major
equipment and training and predicts over 350% ROI in 5 years
obstacle.22
for common platform based virtualization.
SDN’s early adoption in enterprise networks is on its
In [101], a techno-economic model is developed for SDN
way in big companies, such as Google, Amazon, Facebook,
based mobile networks for CAPEX and OPEX savings and
Microsoft, AT&T, etc.,23 A clear business case is present in all
TCO (Total Cost of Ownership). The models predicts 68%
these circumstances, i.e., distributed organizations with many
CAPEX reduction, 63% OPEX reduction and 69% TCO
branches can benefit greatly by software and cloud based WAN
reduction when SDN is adopted for mobile networks in
(Wide Area Network) to provide low-cost Internet services
comparison to the traditional networks. For base-station virtu-
easily and quickly.
alization, [101] assumes CloudRAN architecture but did not
Moreover, there is uncertainty around the benefits in terms
consider the cost of fiber for fronthaul. Similar reductions
of CAPEX and OPEX savings claimed by SDN technology.
are reported by ACG Research report [102] with 68% lower
Virtualization, use of GPP, and resource pooling in cloud com-
CAPEX and 67% lower OPEX by virtualizing EPC. Again,
puting certainly result in cost savings but it is also discussed
the prohibitive cost of fronthaul is not considered here as well.
that cost reduction does not bring the true value of SDN in the
2) Use of GPP: The use of GPP or commodity processors
picture, rather it is the faster roll-out capability with overall
instead of special purpose hardware also reduces the cost sig-
streamlining of network operations which will provide the eco-
nificantly [1]. But this cost saving is based on the view that
nomic benefits to operators and network owners. According to
traditional router vendor equipment is over-priced, although,
a recent survey,24 the flexibility and scalability of the network
not everyone in the industry shares this view [1]. Moreover,
with respect to the demand are the main drivers behind invest-
considering the overall cost of re-organization of telco com-
ments in 5G technologies. SDN and NFV are the key enablers
panies to suit the SDN model, the savings may not be as great
of a network that flex with the traffic.
as predicted. Similar concerns are also raised about the OPEX
The fast-fail approach of SDN will result in quicker roll-out
savings. The challenge of operating a network with white-box
of new services but at this stage, it is unclear what services will
switches from one vendor controlled by a virtual machine from
be commercially viable and what revenues they will gener-
another vendor may be significant and OPEX may be higher
ate [1]. Revenue is a key factor in recovering deployment costs
than the current technologies [1].
and the Return On Investment (ROI) for SDN is an important
3) Resource Pooling in Cloud: Recently, there are attempts
consideration. Although, the faster roll-out of services which
to quantify the cost savings using cloud-based networking.
can reduce the week-long processes to a few minutes may
An initial effort is made in [103] where the Cisco price list
result in multi-million dollar revenue stream for the carrier for
is used to determine the CAPEX of two use cases of mobile
services where time to market is a significant success factor.
networks. The classical networks are compared against a hypo-
1) Virtualization: An NFV approach can reduce the num-
thetical SDN-based network with similar configuration and
ber of physical servers, but NFV is not dependent on SDN
another version of a SDN-based network with virtualization
and although SDN supports NFV by providing a flexible
and sharing. It is shown that a SDN mobile network can save
connectivity platform, virtualization can be achieved without
13.81% CAPEX and with virtualization, the savings can go
it [1]. Service orchestration without any change in infrastruc-
up to 48.04%. The paper, however, made certain assumptions,
ture could generate more benefits than SDN/NFV in the short
such as, the Cisco switches can be upgraded to have OpenFlow
run and a massive transformation of infrastructure can slow
which is not impossible. Although, similar assumptions for
the ramp of savings.25 On the other hand, the SDN integrated
base-stations are not true. We still do not have a concept for
NFV is able to support a sustainable business model for the
a SDN-enabled base-station. The costs of fronthaul and back-
long term future whereas the traditional custom stack based
haul are not considered in the study, and can be significant.
virtualization approaches will likely fail [100]. According
Moreover, the study did not consider the associated cost of
22 http://www.lightreading.com/carrier-sdn/sdn-architectures/defining-use- the disruptive nature of network transformation along with
cases-and-business-cases-for-sdn/a/d-id/716315 training of the staff [103]. In a follow-up presentation [104],
23 http://searchsdn.techtarget.com/answer/Whats-the-status-of-SDN-
the author discusses qualitative comparisons for OPEX sav-
deployments-in-the-enterprise
24 https://www.sdxcentral.com/articles/news/carriers-5g-plans-rooted- ings considering the centralized and simpler control of SDN
sdnnfv-says-ixia-survey/2017/09/ networks, though, concern is expressed regarding the initial
25 http://blog.cimicorp.com/?p=2716 rise in OPEX to get the new infrastructure to work.
3240 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

Fig. 8. CAPEX and OPEX saving estimates when SDN is used with/without virtualization and/or cloudification. CAPEX and OPEX saving estimates for
CloudRAN (CRAN) are also plotted. The reference (100% cost) is traditional LTE network with distributed RAN. Note: details of every estimate method,
assumptions, and limitations are discussed in the text.

Reference [105] performs simulations using the OPNET SDN. However, if cost savings can be translated through to
simulator to compare baseband processing resources for profit, they would represent a significant rise in profit, which
CloudRAN with distributed RAN. The paper concludes that could act as a deployment incentive.
CloudRAN needs 4 times less resources than distributed RAN Another recent study [109] tries to find the central loca-
based on the daily load forecasts. The cost of fronthaul is tion and needed equipment for CloudRAN as a minimization
not considered in this paper which could be prohibitive for problem from CAPEX and OPEX perspective. The results
various operational scenarios. In a follow-up paper, however, show that maximum or full centralization yields a minimum
the authors provide detailed evaluation of split function fron- CAPEX solution for certain LTE-A configurations (40MHz).
thaul cost savings in terms of different multiplexing gains [39]. But interestingly, lower levels of centralization, i.e., more cen-
As expected, the fully centralized solution with all BBU tral locations with small IP/MPLS equipment, also yields up
functionalities in the cloud gives the best multiplexing gains to 18% savings compared to fully centralization when higher
but this puts a lot of capacity requirements on the fron- capacity CPRI is available to cater for high rate LTE-A data
thaul and can only be feasible for operators with cheap (100MHz). The main reason behind this result is the price
fronthaul access. For low traffic load, the BBU pool should gap between big routers (6.72Tb/s and 48 slots) versus small
have higher layer processing and requirements on the fron- routers (1.40Tb/s and 10 slots). Similarly, OPEX savings of up
thaul can be relaxed [39]. The evaluation work in [106], to 7% is shown compared to fully centralized solution mostly
shows a potential of a 30% reduced requirement for resources due to much higher power consumption of big routers in com-
when base-stations opt for centralized processing versus the parison to the small routers. The lower level centralization is
local processing. In the local processing option, each base- shown to save 37% CAPEX and 82% OPEX when compared
station is provided with the resources to handle peak time against fully distributed approach, i.e., the current cellular
traffic. model. This analysis assumes the availability of fronthaul and
The theoretical framework introduced in [107] shows 10- the cost calculations only consider the BBU equipment at the
15% savings in capital expenditure per square kilometer central locations.
when cloud-based RAN is compared against a traditional The fronthaul factor for the cost is considered in the cost
LTE network. However, the theoretical framework lacks some model for CloudRAN developed in [110]. The paper considers
important information, such as, the cost of base-stations in the ratio of cost of fiber per km to the cost of one BBU. When
cloud, and assumes that this cost will be lower than tradi- this ratio is greater than unity, it is not beneficial to go for a
tional systems. The traffic load, usage pattern, and revenue centralized RAN. The paper also showed better results in case
levels in 2015 from an average Finnish network is chosen as of partial centralized RAN solution when cost is minimized
a reference case in [108] to compare the CAPEX and OPEX with respect to all the factors affecting it [110].
of SDN-based LTE versus regular LTE. It is shown that SDN A summary of cost saving estimates are plotted in Fig. 8.
reduces the network related annual CAPEX by 7.72% and Although, the estimates depend on specific method and
OPEX by 0.31% compared to non-SDN LTE. The savings are assumptions used in the calculations. The numbers should not
very small comparing to the annual cost of a MNO (Mobile be compared in strict sense or interpreted as precise bene-
Network Operator) and do not present a compelling case for fits. But, it is interesting to see that virtualization seems to
ZAIDI et al.: WILL SDN BE PART OF 5G? 3241

have a better potential for cost savings than cloudification. It a portion of the network resources for specific applica-
is not surprising, however, given the potential for end-to-end tion, such as, safety. Trials to integrate ProGRAN in
resource sharing with virtualization. Another factor affect- Telefonica’s virtualization project are already underway.
ing cloudification savings is the difference in prices of big 6) AT&T’s FlexWare, previously called Network on-
and small routers [109] and it may not be cost-effective to demand, is based on AT&T’s integrated cloud platform
replace multiple small routers with one big router. Moreover, for service orchestration. It uses SDN and NFV to pro-
an optimal design should also consider the cost of fronthaul. vide software control of different functions for enterprise
4) Products and Business Analysis: As the research in customers and can be used to offer MEC-capable ser-
SDN-based mobile networks is a step behind wired network vices. FlexWare is part of AT&T’s 5G trials which will
developments, there are only a few wireless products emerging also use mmWave spectrum along with WiFi to provide
in the market whereas there are relatively more WAN-related blanket coverage.33
SDN solutions and products. According to SDxCentral,26 the 7) BigSwitch’s Big Mon (Big Monitoring Fabric) can
following mobile network related products are announced evaluate the traffic of millions of mobile subscribers
already: allowing service providers to monitor their network
1) NEC’s NFV CloudRAN27 is a software application performance and ensure ultra-high data performance.34
which runs on the central digital unit (DU), with COTS 8) JMA’s XRAN JMA has recently announced their all
servers, but can delegate L1/L2 functionalities to the software eNB, called XRAN solution.35 They claimed
radio units (RUs) according to the fronthaul condition. to implement layer 1 without hardware accelerators.
It can work with both Ethernet and CPRI as well as one There are a number of products and solutions in wired
DU can have flexible functional split in different parts network domain. Some are relevant to mobile networking as
of the network. well. It is impossible for us to mention all products because
2) ASOCS’s virtual Base-station28 is a fully virtualized of space constraints but some major developments are:
base-station with software of all layers running on GPP 1) Nokia’s Network Service Platform (NSP) is purpose-
servers. It supports flexible baseband partitioning and built Carrier SDN software for service automa-
provides interfaces for digital DAS (Distributed Antenna tion, network optimization and dynamic assurance
System). With ease of installation and integration with for delivery of profitable, on-demand network ser-
any carrier, it offers exceptional total cost of ownership. vices.36 Integrated with Nokia’s Deepfield, NSP can
3) Altiostars vRAN solution29 connects the intelligent perform real-time analytics for resource optimization
RRU with the virtualized compute nodes over any trans- and network security.
port network. The intelligent RRU understands appli- 2) IBM has range of products available in SDN family
cations and schedule packets over the air to achieve designed to create a unified network architecture which
required QoE (Quality of Experience). This may allow enables cloud and big data analytics and optimizes the
full CloudRAN experience even with constraint fron- entire computing infrastructure, i.e., compute, storage
thaul. Initial trials with SK Telecom have been con- and network resources.37
cluded already where key LTE features are validated. 3) An earlier example is Juniper’s High IQ, a pro-
Major customers include SK Telecom, Dali Wireless, grammable networking approach, introduced in 2014.
etc. Dali Wireless has recently announced the devel- ACG Research analyzed three use cases for High IQ
opment of a patented virtual fronthaul interface as an from simplifying CPE (Customer Premise Equipment),
intelligent aggregator-router. That interface makes it pos- and pushing the VPN and firewall functions to the cloud,
sible for multiple operators to connect to multiple BBUs, to real-time network self-optimization and elastic traffic.
creating a multipoint-to-multipoint network.30 The report indicated compelling business cases through
4) Advantech’s Packetarium XLc carrier grade glade significant CAPEX and TCO savings in all the use
server31 is designed for NFV infrastructure for cases.38
CloudRAN and mobile edge computing (MEC). Its can 4) In 2016, ACG Research’s white paper compared HPE
host 9-12 Intel Xeon processor based blades and sup- pre-integrated NFV solution to a bottom-up DIY solu-
ports 720Gbps I/0 and more than 1.2Tbps switching tion and show the extra OPEX savings of 30% can be
capacity. achieved with HPE pre-integrated solution only.39
5) Netsias vRAN platform, called ProGRAN,32 lets 5) Lessons Learned: From this review there is generally
network operators slice the RAN and effectively allocate consensus that a positive business case for SDN is evident.
26 https://www.sdxcentral.com 33 https://www.sdxcentral.com/articles/news/att-will-use-flexware-platform-
27 https://www.sdxcentral.com/products/nfv-c-ran/ waco-5g-trial/2017/12/
28 https://www.sdxcentral.com/products/asocs-virtual-base-station-vbs/ 34 https://www.sdxcentral.com/articles/news/big-switchs-big-mon-keeps-
29 https://www.sdxcentral.com/products/altiostar-vran-solution/ eye-mobile-subs/2017/12/
30 https://www.sdxcentral.com/articles/news/dali-wireless-uses-sdn-to- 35 http://www.jmawireless.com/products/xran
36 https://networks.nokia.com/products/network-services-platform
virtualize-the-fronthaul/2017/10/
31 https://www.sdxcentral.com/products/packetarium-xlc-carrier-grade- 37 https://www.sdxcentral.com/listings/ibm/
blade-server-for-virtual-service-edge/ 38 http://www.juniper.net/us/en/local/pdf/whitepapers/2000581-en.pdf
32 https://www.sdxcentral.com/articles/news/netsias-platform-lets-operators- 39 https://www.hpe.com/h20195/V2/Getdocument.aspx?docname=4AA6-
slice-dice-ran/2017/12/ 4924ENW
3242 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

A significant issue of network transformation costs is identi- the tech-ready state for production deployment. The testbeds
fied, based on the disruption to operational approaches and available for implementation are also limited by GPP con-
networking infrastructure. The radical change in the com- straints as discussed in Section III-B and until we have the
munication network architecture, with full softwarization and platforms suitable for signal processing, it is very difficult to
centralization, focusing only on the broader CAPEX and see a real and comprehensive SDN-based mobile network. We
OPEX savings in future does not account for transformation have also listed the outstanding issues and problems related to
costs. A rapid transformation model is very unlikely to be SDN-based mobile network architecture.
followed, due to the high risk of service disruption. The rest of the section is mainly organized in terms of SDN
A more practical alternative would be an incremental or planes and what has been proposed for each of them and what
evolutionary approach to the introduction of SDN into the ideas are repeatedly discussed among the research community.
operational networks. Although, there still need to be indi- Table V shows the summarized landscape and an overview of
vidual business drivers or short-term monetization for each the repeated themes, such as, the realization of EPC nodes as
incremental implementation to make it appealing for carri- applications in SDN-based mobile network, distributed control
ers and network owners [1]. While the evolutionary path to plane, and data plane comprising of CloudRAN and hetero-
full SDN deployment is also not clear, the individual busi- geneous radio technologies- e.g., LTE, 3G, and WiFi, etc. for
ness cases for the increments are among the big unknowns. backward compatibility- and OpenFlow (OF) switches. The
Wherever, an enterprise has found some compelling cases to themes are grouped under the associated SDN layer in Table V.
upgrade, the SDN point solutions happen [1]. Carriers will
establish a roadmap of SDN/NFV solutions which will min-
imise disruption and show short term business benefits [1]. A. Application Plane
Thus, to determine the duration of transformation to a full While transforming LTE architecture into SDN’s 3 plane
production deployment of SDN is challenging. model, EPC control nodes, e.g., MME, HSS, and PCRF are
usually suggested to be implemented as application plane
modules as shown in Table V [4], [63], [90], [122], [123],
IV. C URRENT S TATE OF THE A RT OF SDN-BASED [125], [126], [128]. Similar ideas are evident into the evolv-
M OBILE N ETWORK A RCHITECTURE ing 5G system’s architecture from 3GPP [2], where mobility
As we have seen in the previous section that there exists a management and policy control functions are connected to
gap between the state where the SDN-based mobile network other new functions, such as, authentication, application man-
technology will be ready to be deployed and the current state agement, etc. via a common message bus. The architecture
of the solutions. But not all aspects of the SDN-based mobile is a step closer to the softwarization or cloudification of
network architectural development are covered in the previous core network. In NFV enabled environment, the core network
section concerning major challenges. In order to really appre- modules can be implemented as VNFs (Virtual Network
ciate the current state of the art, we should also look at the Functions [135], [161].
major body of relevant literature. A detailed discussion about However, the case with S/P-GWs is not very clear. In the
all the work under SDN-based mobile network architectural 5G system architecture from 3GPP, the User Plane Function
development is outside the scope of the current paper. In this (UPF) to handle the user plane path of PDU sessions remains
section, we have discussed the major lessons from our study in data plane [2]. UPF in 5G architecture is supposed to the
in the form of common and popular themes and approaches equivalent of PGW-U (User plane part of PGW) in EPC. The
which may become part of the standards. equivalent of PGW-C (Control plane part of PGW) is SMF
The earlier papers on SDN-based architectures talk mostly (Session Management Function) in 5G NGC (New Generation
about the benefits as they have to make a case for SDN Core) [2]. The separation of control and user plane in 5G NGC
technology. The architectures they presented are mostly very is more comprehensive than in EPC. Although, this separation
generic and high-level. As time goes by, the proposed architec- is not exactly the same as the separation of control logic and
ture still remains high-level but there are publications which forwarding hardware defined in SDN.
are more focused on the issues and emphasize on details. A detailed study about which functions of LTE EPC should
Recently, there has been a lot of work on specific network be moved to cloud in an SDN based system is done in [129].
services, such as mobility management [111], resource They pointed out that as functional blocks move to the cloud,
management and sharing [112]–[114], CoMP (Coordinated the cost of S/P-GWs goes down as the gateway hardware
Multi Point) [115], service chaining [116], network slic- becomes simpler but data overhead and end-to-end latency
ing [117], [118], service-customized networking [119], energy increase. The straight forward solution of moving all control
efficient traffic engineering- e.g., optimized routing, switching functionalities to the cloud may result in unacceptable latency
and power management [120], integration with context-aware for some applications. Also, cloud infrastructure performance
networking architecture [121]- etc. In this paper, we stayed will be a critical factor for high frequency of control plane
away from network services as a thorough coverage would operations taking place at S/P-GWs.
make this paper too long. On the other hand, moving signaling control and resource
There has been preliminary implementations with some management logic to the cloud would allow S/P-GWs to
basic results regarding overheads, cost savings, etc., but a be deployed on distributed elements but would also create
big gap still exists between the current state of the art and bottleneck at the centralized control when forwarding rules
ZAIDI et al.: WILL SDN BE PART OF 5G? 3243

TABLE V
S UMMARY OF R ESEARCH IN SDN-BASED M OBILE N ETWORK A RCHITECTURES

are exchanged. Moving resource management back to the also suggested a hybrid approach where the scenario and traffic
data plane would make it more independent and resilient requirements would dictate the design but this approach would
but it would not be possible to take the advantage of require state synchronization and orchestration between cloud
centralization through optimal resource allocation. The authors and data plane to avoid redundant assignments.
3244 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

A comparative cost analysis of SDN-based EPC and soft- A distributed controller architecture is presented in [131],
ware only EPC, where LTE control and user (C/U) plane are where controller-domains are defined as the eNodeBs served
both implemented in cloud, is presented in [127]. The key find- by a particular controller. Handover processes are also defined
ing of this study is the realization that a pure software solution between eNodeBs in a controller domain and within different
where C/U planes are in the cloud is much more expensive controller domains. The handover processes and packet sizes
and it is economical to leave the user plane (S/P-GWs) out- are also revised from 3GPP standards to reduce the signaling
side the cloud and perhaps on the dedicated hardware. An overhead. The proof-of-concept, i.e., testbed implementation
optimal solution for placing SGW-C- i.e.,- the controller part or simulation experiments are missing from this study to ascer-
of S-GW, in the network is presented in [124] using game tain that the design is complete and adequate for 5G capacity
theoretic approaches. The algorithm finds a trade-off between and speeds. The controller domain idea makes the SDN control
load reduction on SGW-C and reducing relocation of S-GW plane scalable but important details about controller place-
which is costly for the mobile operators. ment and optimization are also not available in this paper.
In [122], SGW-C, i.e., the control part of S-GW is used as Optimal controller placement is also an open issue. Solutions
an application along with MME module and PGW-C (control are presented in [124], [134], and [135] using optimization
part of P-GW) with an OpenFlow controller. The data plane, or function, linear programming, and game theoretic approaches.
SGW-D, is comprised of advanced OpenFlow based switches In [135], the optimization is done over the allocation of VNFs
with GTP support. The numerical evaluation show reduction (Virtual Network Functions) and controllers to the data cen-
in signaling load with OpenFlow implementation of EPC than ters along with placement of data centers and SDN enabled
the original 3GPP EPC. This work was extended in [123], switches for data plane. The objective is to optimize network
where all functional blocks of EPC including S/P-GW control load and data center resources.
part are implemented as application over mobile controller. SoftRAN [3] also observes that it is cost-effective to leave
The signaling load in full implementation of EPC is shown to data plane functionality to the base-stations along with some
be lower than that of [122]. part of control plane for delay-sensitive decisions, but stressed
on the coordination of closely-deployed BS in a dense network
though a centralized controller such that the dense base-
B. Control Plane stations in an area can be viewed as virtual big base-station
One of the important issues in SDN-based networks is con- (big BS).
troller scalability. Clustering or distributed control plane is Similar ideas of a regional controller which possibly resides
explored as a potential solution, although, many important in cloud and local controller, located in a macro cell, have
questions, such as, inter-controller coordination, etc., remain been used in [136]. The SDN-based mobile network model
unanswered. Hierarchical control planes are also proposed to for dense deployments [136] is an outcome of FP7 project
cater for the time-sensitive control processes on one hand and CROWD with the focus on energy optimization. The paper
the need for centralized view for optimal operations on the also proposed a mobility management scheme for their model.
other. The architectures presented in [137] and [139] defines a
1) Hierarchical/Distributed Control Plane: Different local SDN controller (LSC) for scalability which are respon-
proposals are put forward which define domains or sible for heterogeneous wireless networks in an area and are
clusters for each controller to resolve the scalability connected to core SDN controllers (CSC). The paper [139]
issue [90], [126], [130]–[132], [137]–[139]. The controllers goes into the details of LSC and what functions it is sup-
can be connected to each other in a distributed manner posed to perform including content caching, resource and
or there could be a higher level controller to coordinate mobility management functions. The authors also showed the
between the lower level controllers. The size of the domain backhaul traffic reduction with content caching via simulation
or clusters can be chosen to satisfy the delay constraints. experiments. The virtualization layer lies on top of physical
Hierarchical control plane proposals suggest 2 or 3-tier infrastructure and each VM (Virtual Machine) communicate
architectures for scalability [126], ease in management to LSC via a local SDN agent. The implementation results
specially in heterogeneous networks [137], and for fulfilling are still preliminary and does not show the interworking of
delay requirements [130], [137]. different LSCs and does not include offloading mechanisms
A recent survey of the issues and challenges in distributed over heterogeneous technologies.
control is presented in [133]. The paper identifies and surveys Reference [90] proposed a three-level control plane structure
possible options for controller deployment including central- for SDN-based mobile network. They have included device
ized, physically distributed but logically centralized control controller (for UE), an edge controller, implementing 5G com-
plane, i.e., flat or hierarchal. The available implementation plaint network functions through control applications, and
of controllers are also classified in the two groups [133]. an orchestration controller to coordinate utilization of cloud
However, there is no standardization of controller-controller resources. Two versions of edge controller are identified in
interface yet, which is the major challenge in realizing a dis- the paper, one implemented in the cloud and the other in the
tributed control plane architecture. Efficient network monitor- mobile device for mission-critical control and out-of-coverage
ing, maintaining consistent network state, and interoperability services. The paper does not go in the detail of physical
of equipment from different operators and manufacturers are structure or fronthaul or backhaul issues rather points out the
also big issues in a distributed environment. requirements for the control plane architecture. Following up
ZAIDI et al.: WILL SDN BE PART OF 5G? 3245

from [90], Trivisonno et al. [138] define the initial attachment resource utilization solution can provide immediate moneti-
and service requests procedure for the 3-level control plane zation opportunity.
structure. They show that reconfigurability of the network to In [69], a high-level heterogeneous network architecture is
use appropriate cloud resources can improve the latency up to presented with a SDN controller providing network control
75% compared to 3GPP Release 12 complaint 4G systems. functionalities to RAN, transport network, service providers,
Reference [132] describes a high-level architecture for virtual operators, etc. The paper did not go into the details
mobile networks with one controller each for core network of implementations or low-level architecture of the controller
and RAN. They use a coordinator to provide end-to-end but discusses high-level requirements from different interfaces.
services through both controllers. Some services, such as, The discussion in [172] also remains at high-level in describing
firewall, encryption, etc., are implemented through middle- the scenarios and benefits of using SDN to control and operate
boxes and the controllers decide the flows traversing through a HetNet.
these middle-boxes. The paper identifies major challenges The approach discussed in [9] proposes a SDN controller
of this approach, such as, providing fine-grained con- cloud providing control to data plane including heterogeneous
trol which is not yet possible with OpenFlow. Moreover, wireless access systems and BBU cloud for CloudRAN. The
the paper provides preliminary performance results through SDN controller will provide control to the EPC cloud imple-
simulations. mented in GPP as well as physical switches for the core
2) Global Network View (GNV): The paper [140] intro- network. The paper provides details about different compo-
duces the concept of global network view (GNV) which nents of data plane and how they can be specifically controlled
stores information about network state and can be accessed by with associated issues and benefits. However, the paper still
network services. The storage can be local for delay-sensitive remains a high-level conceptual contribution.
information, such as, channel state information. The rest of the HetSDN [142] also proposed seamless transfer between het-
information regarding flows and load can be stored in a com- erogeneous RANs via virtual switch. They introduced two
mon place. The protocol stack in the architecture of [140] is new components, i.e., SDN-NS (SDN Network Selection) and
in the modular form which can be orchestrated by the control SDN-HA (SDN Home Agent). SDN-NS uses a virtual switch
plane for implementation of specific function, e.g., eNodeB or to seamlessly direct packets to the most suitable wireless
S-GW. interface and SDN-HA is used as the anchor for all traf-
The use of GNV with locally and globally stored infor- fic for the UE. Another idea to provide seamless movement
mation is also discussed in [141] but only for the context of between LTE and WiFi is through SDA (Software Defined
access network. A similar idea is introduced in [128], where Access) [173]. In this architecture, a separate controller, i.e.,
an additional plane called ’knowledge plane’ is added on top SDA controller is added on the control plane which has a
of the application plane in SDN-based mobile network archi- client side SDA-u on the UE. The also have another element
tecture. This knowledge plane is responsible for providing a SDA-g (SDA gateway) on the data path which is used as an
global view of network usage which can be used to device anchor. Prototype implementation shows proof of design and
new applications to optimize resource utilization. The useful concept.
information in the knowledge plane consists of network usage, VCell [143] describes an architecture with heterogeneous
traffic load, congestion information, etc. cells, i.e., macro, micro, femto, etc., where all resources
3) HetNets: The current mobile network environment, in are virtualized in a logically centralized pool. The resources
general, consists of heterogeneous radio technologies, such are allocated to the user from a 3D grid of time, space,
as, 2G/3G/4G and WLAN. Most likely, the future operat- and frequency resource blocks. The SDN controller assigns
ing scenarios will also have similar characteristics. SDN has resources and manages interference. The main contribution
the potential of exploiting the different radio resources in of this paper is to consider heterogeneous cell types while
a comprehensive manner as the details of the technology building up the resource grid. The issues with centralization,
are hidden under the abstraction layer and it is much easier such as, fronthaul latency and capacity, processing latency of
to optimize the utilization of available resources in a uni- commodity hardware, etc., are not discussed in [143].
fied way [137]. Although, the mobility management, joint Heterogeneous CloudRANs are also studied in literature
resource allocation, end-to-end QoS (Quality of Service) assur- where small RRUs and macro RRUs are distinguished by their
ance, and security issues pose additional challenges specially service area [151]. These networks may also contain small or
when UE moves between networks of different ISPs using macro base-stations based on the availability of resources and
heterogeneous technologies [144]. chosen functional split. The macro base-station or macro RRU
Special network selection switches are proposed to achieve is shown to provide signaling for the whole service area and
handover between different technologies [142] and virtualiza- small base-stations and RRUs can only be used on-demand
tion is also used as a mean to connect different networks optimizing energy efficiency while satisfying QoS constraints.
to a controller [139], [143]. Traffic offloading schemes are
also coming up recently [145]. In [112], a resource alloca-
tion mechanism with holistic view is introduced for SDN- C. Data Plane
based heterogeneous networks. Most of the work in this CloudRAN seems to be preferred choice for SDN-based
respect remains at conceptual level and more exploration mobile network layout, although, it is not always benefi-
is needed [9], [69], [172]. This is one area where optimal cial to centralize all the processing in the cloud [106], [143].
3246 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

The case for partial centralization of processing units is dis- The design in [158] put all the EPC control functions in con-
cussed under various themes, such as, in RANaaS (RAN troller which translates the flow rules for OpenFlow switches
as a Service) the level of centralization is chosen accord- used in place of S-GW. P-GW is eliminated from the design.
ing to the need and network state [73], [147]. A plastic or The author claims efficient mobility support with the proposed
adaptive architecture which provides various levels of central- EPC architecture than the legacy one. The paper overlooked
ization according to the operational requirements [148] could the real problem of scalability and translation of huge lists of
be very desirable from business point of view as it provide policies into simple flow rules for the switch [66]. Until both
further optimization of cost but the technical issues involved problem remain unsolved, it would be difficult to realize the
in realizing such architecture are not clear at this point. architecture.
Reference [149] discusses the partial centralization option Reference [64] added relevant extensions in the OpenFlow
in their architecture to cater for the fronthaul and backhaul protocol for GTP tunneling and encapsulation. In [156], a new
constraints. dynamic GPRS tunneling protocol is presented and evaluated
Reference [144] pointed out that, in a CloudRAN integrated for SDN based core network. The protocol allows elastic use
HetNet, if multiple standards are being used in the same spec- of data plane resources and enable cloud to provide on-demand
trum, the RRUs can support them only partially. Whenever packet processing. Terminating GTP tunnel at the cloud would
the standard changes, the BBU is forced to restart instead of incur additional delays than terminating it on the fast data
sharing multi-standards resource directly. path. The paper is preliminary work and more evaluation is
Reference [152] suggests an architecture for integrating needed to understand the scalability and performance with real
existing and new fronthaul and backhaul networks into a networks.
flexible unified 5G transport solution using SDN/NFV-based The integration of OpenFlow in EPC is discussed in [157],
management and orchestration (MANO) technology. The where the S-GW is replaced with OpenFlow switches and
paper explores the resource management functional blocks and detailed steps are discussed to keep GTP tunneling and ver-
discusses some use cases. tical protocol stack for backward compatibility. The paper
Issues and advantages related to the integration of SDR later introduced different processes like creating, modifying,
(Software Defined Radio) with SDN are also explored in [153] and deleting a bearer or a session without GTP tunneling for
and [154]. Preliminary results have shown that the integrated lightweight architecture.
network with the potential of using spectrum holes and white In [155], OpenFlow 1.4.0 is extended and optimized for
spaces performs significantly well than the network without 3GPP EPC including the design of controller, switch, and pro-
SDR [153]. tocol. The paper first highlights the requirements including the
1) Exclusively OpenFlow Based: OpenFlow seems to be clear definition of northbound APIs, GTP support, latency, etc.
the choice of the researchers at least for now [4], [8], [64], The switch is evaluated within OpenEPC implementation for
[71], [113], [116], [122], [126], [132], [155], [157], [158]. proof-of-concept and scalability.
OpenFlow and SDN may even appear to be synonyms in some CellSDN [4] identifies additional requirements from
places. During our study, we only came across one paper [165] OpenFlow switches to provide network services, such as,
which advocated the use of ForCES (IETF Forwarding and deep packet inspection, header compression, etc. They
Control Element Separation framework) instead of OpenFlow. emphasized on reducing the controller’s load by means of
ForCES includes abstraction model for both SDN and NFV local control agents at the base-station and switches. The
as described in RFC 5812 [165]. The major issues in using interfaces between controller, local agents, and data plane
OpenFlow for the data plane in a cellular network are the are also defined in terms of requirements and available
lack of GTP tunneling support and the transformation of com- protocols.
plex billing and management policies into simple and scalable 2) Control Signaling and Data Split: The approach called
routing rules for OpenFlow switches. CONCERT [11] separates control signaling from data trans-
Among the earliest work on SDN-based mobile network, mission as defined in BCG2 architecture and Phantom cell
OpenRoads or OpenFlow Wireless presents a prototype concept (see Section II-C). It uses SDN switches with all
where wireless APs (WiFi and WiMax) are augmented radio interfaces, computational resources, and servers. These
with OpenFlow [174]. A centralized controller, Network OS switches are then controlled by the control plane entity
(NOX), is used to host the application plug-ins and trans- called a conductor. All management and virtualization ser-
late the policies into routing rules for OpenFlow. Applications vices are defined as control plane modules. Application
of mobility management were executed over the prototype. scenarios are highlighted to focus on the benefits of
Authors suggested similar OpenFlow additions for LTE, SDN-based architectures. Technical issues are also briefly
although, LTE base-station and core network is way more explored.
complex than a WiFi or WiMax access point and it is not Artuso et al. [46] also touched upon the control-data split
straightforward to translate management and billing policies in their deployment scenarios but did not go into the details
into simple routing rules [66]. Softcell [66] remarked that the of implementation. [151] also discussed the idea in relation
policies of a cellular network translates into thousands of rules to heterogeneous CloudRANs, where macro base-station or
and it is important to come up with some logical aggrega- RRU can provide signaling to the area and small base-stations
tion techniques so the rules can be fit into today’s routers and or RRUs can only be used on-demand. The switching-off
switches. strategies are also discussed in [151].
ZAIDI et al.: WILL SDN BE PART OF 5G? 3247

2) NFV in SDN Domain: As the SDN architectural propos-

als are concerned, NFV integrated SDN architectures make
more sense from business as well as technical points of
view [4], [8], [11], [71], [139], [154]. Designs may include the
role of hypervisors [4], [8], [164], [165] or virtualization layer
above the infrastructure or forwarding plane [4], [139], [166].
In CONCERT [11], the virtualization and management ser-
vices are provided as control plane modules. These architec-
tures are high-level designs and the relevant issues are not
fully explored.
One of the initial attempts to combine NFV and SDN tech-
nologies together for mobile EPC is given in [163]. It shows
the complexities and issues involved in achieving the vir-
tualization capacity in a softwarized core network. Firstly,
the authors show the use of OpenFlow switch network to
Fig. 9. Possible options of positioning SDN Resources, SDN Controller and replace S/P-GWs of EPC. Important questions left unanswered
SDN Applications in NFV Architectural Framework [161]. here, such as, how the huge lists of policies in PCRF can
be translated into manageable flow rules for the OpenFlow
switches. Secondly, the combined NFV and SDN architec-
D. Integrated Virtualization ture is proposed which shows two different instantiations of
1) SDN From NFV Perspective: As mentioned earlier SDN the control plane, i.e., one for NFV and the another for end-
and NFV are complimentary technology enablers for 5G to-end connectivity. The two instantiations are connected to
and [161] provides an overview of both these technologies each other and also to the NFV management and orchestra-
in the context of 5G network requirements and architec- tion unit, at supposedly higher control level. The EPC control
ture. From NFV perspective there are different options for plane is also mapped as VNF (Virtual Network Functions) in
deploying SDN within the NFV infrastructure and the var- the NFV infrastructure. Many issues are unanswered regarding
ious deployment options have been analyzed in detail by the design. Most importantly, how these VNFs will interact
ETSI NFV in [160]. Figure 9 [161] summarizes the possi- with SDN controller to maintain connectivity. Scalability is
ble options of integrating SDN application, SDN resources another issue which needs to be tackled in a centralized EPC
and SDN controller with different entities within the NFV architecture. A layered approach providing clear demarcation
MANO and NFV architecture. There are five integration between actions needed for connectivity and providing ser-
options for SDN controller to either (i) be part of OSS/BSS vices is probably a better way to understand the problem
(Operations/Business Support Systems), (ii) exist as an entity instead of modeling a broad set of NFV and SDN aspect into
within NFVI (NFV Infrastructure), (iii) exist as a PNF one flat design as also mentioned in [163].
(Physical Network Function), (iv) be instantiated as a VNF SoftAir [8] proposed an SDN-based mobile network archi-
(Virtual Network Function), or (v) be integrated within the tecture with network hypervisor to orchestrate virtualization.
VIM (Virtualised Infrastructure Manager). The latter approach, They have proposed wireless hypervisor and switch hypervi-
for example, is supported by the ONF SDN architecture [159] sor for SDN enabled base-station and SDN enabled switches.
and is also an adopted approach by open source OPNFV They have designed SDN enabled base-station such that all
platform.40 OPNFV prescribes the integration of SDN con- processing stays at the cloud except modulation/demodulation
trollers like ODL (OpenDayLight) and ONOS (Open Network which are located with antennas at RRUs. The split would
Operating System) with OpenStack VMM (Virtual Machine reduce the load on CPRI supported fronthaul.
Manager). It should be noted that from ETSI NFV perspective, The paper [165] advocates the use of ForCES (IETF
OpenStack platform for all practical purposes is considered as Forwarding and Control Element Separation framework)
a VIM platform. instead of OpenFlow. The authors build a proof-of-concept
There are also some prominent research projects like 5G focusing on GTP tunneling. The proof-of-concept involved a
NORMA41 that leverages on the SDN and NFV concepts in hypervisor which instantiate VMs and deploy necessary LFB
order to develop a novel mobile network architecture that shall (Logical Forwarding Block), an element of the forwarding
provide the necessary adaptability in a resource efficient way plane in ForCES. They build an application to provision and
able to handle fluctuations in traffic demand resulting from het- destroy GTP tunnels and collect statistics. The authors claim
erogeneous and dynamically changing service portfolios and that the framework can scale-in and out as required but there
to changing local context. In [119], a framework is presented are not analytical results presented in the paper to support the
which identified the management and operation functions to claim. Proof-of-concept was shown to the IETF meetings but
deliver service-customized networking to various customers in no results are presented in the paper.
a SDN/NFV-enabled ecosystem. A high-level architecture is also presented in [166], where
a network virtualization layer is laid over the physical infras-
40 https://www.opnfv.org tructure and under the service enablement layer which seems
41 https://5gnorma.5g-ppp.eu/ to be very close to control plane in concept. Architecture is
3248 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

using cognitive and cloud optimization domain and is named separation of forwarding and control, transforming tradi-
as CONE architecture. As it is also mentioned in the paper, tional infrastructure to a cloud infrastructure, and transforming
many details are left out of the paper and only preliminary Telecom-oriented systems to Internet-oriented systems. For
design is discussed. RAN, SoftCom is focusing on moving the control plane for
The proposed SDN-based mobile network architecture small cells to macro base-station and centralization of BBU
in [71] moves EPC to cloud to benefit from NFV while processing as in CloudRAN. SoftCom promises to develop
maintaining 3GPP interfaces with legacy nodes. It suggests an eco-system for simplification of network operations and
moving some functionalities of eNodeB, if required, to the innovation of open business models.
cloud but the prototype uses the unchanged eNodeB. The
proposed architecture is deployed over a testbed using two
eNodeB from Nokia, OpenFlow enabled MPLS switch, open E. SDN Applications
source S/P-GW and probing tools. The results show downtime In this section, the discussion about application of SDN in
of up to 2 seconds in live migration of a MME running on a improving and optimizing network management is included
VM instance. only to admire how the SDN research space is expanding and
Mobileflow [63] is one of the earliest solutions to resolve the focus is turning towards details. There is huge amount of work
compatibility issue between SDN-based networks and legacy in this respect specially since 2014, and it is outside the scope
systems using virtualization techniques. OpenRAN [7], [175] of the present paper to provide a comprehensive survey on
is also a high level idea of a SDN-based mobile network applications.
through virtualization. A SDN controller would first create 1) Resource Sharing: A framework for co-primary spec-
virtual RRU and virtual BBU using the physical resources trum sharing among MNOs (Mobile Network Operators) is
and then dynamically optimize them according to the developed in [114]. The framework is based on an SDN
application. architecture integrating Mobileflow [63] and SoftRAN [3].
Reference [176] develops an interesting concept of Hybrid- Co-primary spectrum sharing enables optimal resource utiliza-
eNodeB (HeNB) which hosts a virtualized EPC (vEPC) along tion among different operators. Mobileflow [63] is the core
with all eNodeB functionalities. vEPC is the most fundamental network part of this scheme while SoftRAN [3] is evolved
version of EPC without the complexities of detailed con- to enable access network sharing. The big base-station con-
trol. vEPC needs the connections to physical EPC, although cept of SoftRAN is used in the paper for all base-stations of
not all the time, to get the necessary information about all operators. The logically centralized controller then assigns
billing, roaming, etc. HeNB greatly simplifies the deploy- the resources based on the resource management rules. The
ment of 4G LTE and even allow standalone and self-powered scheme does not go into the details of SLA (Service Level
UAVs (Unmanned Aerial Vehicles) and LAPs (Low Altitude Agreements) between MNOs and does not identify the busi-
Platforms) to host HeNB. Reference [164] presents the idea of ness incentives and benefits. Security issues are also left out
virtualized EPC using a general purpose node (GPN). GPN is a of the paper.
core-class server with a hypervisor to provide virtual instances Radio resource sharing is also discussed in [113], where
of EPC. The main purpose is to reduce cost through improve- virtualization of LTE eNodeB is used to facilitate lease
ment in resource utilization. EPC nodes and interfaces remain on-demand the physical infrastructure and resources of one
unchanged in this work along with GTP tunneling. operator to the flows of another operator. The framework is
In a recent publication [154], the authors presented an archi- named OpeNB (Open eNodeB) and it is based on SDN and
tecture for WNV (Wireless Network Virtualization) which is OpenFlow technologies. Besides OpenFlow controller, another
similar to NFV concept but only for access networks. Their controller called OpeNB controller provides signaling and
architecture places the virtualization layer or hypervisor as management control for resource sharing. A main controller
southbound interface between the physical network compo- provides hand shaking between networks of different opera-
nents or the data plane and the virtual resource management tors. System-level simulations show improved performance in
or the control plane. The virtual network functions are placed terms of packet drop rate and throughput.
in the application plane. The paper also listed the outstanding Another proposal for SDN-based smart gateways to con-
issues and advantages of such an architecture. nect S/P-GW of multiple operators is presented in [45].
The proprietary virtualized access (vAccess) development Here a SDN orchestrator is responsible to manage multiple
platform from Freescale [55] is created to facilitate the devel- smart-gateways in order to provide demand-based uplink
opment of SDN/NFV products. It is built over OpenStack capacity to the connected small cells. The smart-gateways
cloud computing platform with hardware accelerators for L1 allow resource sharing among multiple operators through SDN
real-time processing and Linux patch to support bounded orchestrator.
guarantees for application start time. The vAccess platform 2) Mobility Management: Karimzadeh et al. [111]
provides the basic structure to build LTE base-station software. looked at the requirements of DMM (Distributed Mobility
In 2013, Huawei [146] unveiled SoftCom strategy. SoftCom Management) and how SDN architecture can satisfy them.
is a holistic approach to network architecture based on cloud They also showed a simple example of how DMM can be
computing, SDN, and virtualization. The four key elements of implemented in a virtualized LTE environment. Their follow-
SoftCom are cloudification of equipment through separation up paper also provides simulation results of SDN-based DMM
of hardware and software, cloudification of network through implementation [167]. Their performance evaluation results
ZAIDI et al.: WILL SDN BE PART OF 5G? 3249

show that seamless mobility management is possible when 2) Backward compatibility The non-standalone 5G allows
X2 path is used between eNodeBs. an intermediate phase with backward compatibility to
The benefits of SDN in mobility management with global LTE by enhancing eNB and defining new interfaces
network view and flow-based control are highlighted in [168]. between 5G and 4G entities which can interpret the stan-
The authors discussed mobility management in SDN architec- dard communication procedures of 4G for the context
tures with different centralization levels. They identified issues of 5G entities. GTP tunneling is also preserved in 5G
regarding session continuity and scalability of handover and NR [2].
showed how mobility management can be easily integrated 3) Distributed control Distributed control is very impor-
with autonomic management mechanisms to optimize resource tant for network scalability. So far, OpenFlow does
utilization. not have standardized controller-controller interface and
3) Multi RAT Scenarios: A SDN enabled traffic offload- more work in needed in this regard.
ing scheme is proposed in [145]. A SDN controller runs 4) Global network view Global network view, i.e., a cen-
software based RRM (Radio Resource Management) and tralized source of information, is a major attribute of
PCRF (Policy Charging and Rules Function) modules to deter- SDN-based networks. The associated research explores
mine the offloading from LTE to WiFi depending on the local and centralized storage options for time-sensitive
network state and flow requirements. If both networks are and global information respectively. We are not aware
unable to satisfy traffic requirements, the packets are dropped. of any standardizations activities in this respect.
Simulation results show 15% less packet drop with the SDN- 5) Access control There is a lack of standardization
enabled offloading scheme. RRM is responsible for updating for application access management, over northbound
the network state information and PCRF is responsible for interface, regarding rules for granting privileges, binding
policy enforcement. The paper shows a very important use mechanisms, auditing, trust, etc.
case where SDN can provide performance improvement in a 6) Secure southbound interface Dedicated and secure log-
straight forward manner. ical links between controller and data plane entities
A detailed sub-optimal online solution for end to end routing are very important, though expensive, for sustainable
in mobile networks based on SDN and cloud computing is operation of the network. There are no reported stan-
presented in [169]. The objective is to maximize the amount dardization activity in this regard.
of traffic accepted over time. The problem is formulated and 7) White box deployment white-box solution requires a
an online algorithm is developed and tested via simulations lot of standardization across the hardware and software
for multi RAT (Radio Access Technology) scenario. domains to simplify the deployment and systematize
A wireless/wired backhaul solution based on SDN is the upgrade procedures. Meanwhile we may see overlay
presented in [170] for small cells. The prototype performance technologies on existing infrastructure, such as, hyper-
is compared against 802.11s and significant improvement is visor for virtualization will offer minimal disruption
observed when channel conditions are used to decide which but will not provide flexibility to use the hardware
packets to transmit over the backhaul. Similar idea is also efficiently.
presented in [171] where some details of the on-going work 8) Centralized BBU Although, 5G NR standard’s recent
on implementation are also discussed. An SDN controller can release 15 has not discussed splitting of RAN in
adaptively power on/off small cells based on the demand and RRU/BBU and centralized BBU processing in the
traffic metrics. cloud [2], but historical reasons and saving potential
may very likely push for the CloudRAN layout. If it
does, the standard should include flexible and adaptive
F. Techniques and Standardization models suitable for broad range of operating conditions,
In this section, we summarize the discussion about state e.g., availability or non-availability of fibre, etc.
of the art of SDN-based mobile network research by listing 9) Functional split fronthaul NGFI Chinamobile
some of the most promising techniques we came across and presented NGFI (Next Generation Fronthaul
the associated standardization activities. For some issues, the Interface) [35] which put all antenna related func-
standardization process is quite mature, e.g., NGFI for split tionality, such as, downlink antenna mapping, fast
function fronthaul, while for others, the process is still in Fourier transform [FFT], channel estimation, and equal-
infancy or not started at all. The list is by no means exhaustive. ization in RRU. Currently, NGFI is being standardized
1) Control/user plane split NGC (Next Generation Core) under IEEE 1914.1 project.
of the recent 5G NR (New Radio) non-standalone stan- 10) Ethernet for fronthaul The most important activity
dard is a step towards a fully softwarized core by to mention in this respect is IEEE 802.1CM Time-
complete separation of control and user plane and mod- Sensitive Network task group. The group is looking into
ularizing the management functions. Control/user plane the set of standards focusing on time-synchronized low
distinction in 5G is not equivalent to SDN’s con- latency high bandwidth services over Ethernet networks
trol/forwarding plane definitions and many issues are for fronthaul transport network.
still left towards realization of SDN-based core network. 11) MultiRAT Now a days, the devices working in multi-
Radio access part in 5G NR, gNB, contains both control RAT scenarios have multiple interfaces. In SDN ecosys-
and user plane. tem, devices will be oblivious of the multiRAT paths and
3250 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

a generalized abstract view can be used at the controller 6) GPP implementation EPC control nodes are suggested
level. This will also require standardized generalization to be implemented as applications but carrier grade
and southbound interfaces to various technologies. implementations are needed to ascertain the realizability
12) ETSI’s NFV MANO framework ETSI’s NFV MANO of the idea.
framework is driving the development in NFV domain. 7) Edge computing An opposite concept to centralization,
The connectivity provided to the framework by SDN edge computing also brings promise of performance
network also requires interfaces between MANO frame- enhancement and cost savings. It is vital to bring the
work and SDN controller where connection require- two ideas together to see how they interplay and how
ments and resource assignment information can be practical it is to develop a plastic architecture that could
exchanged. Moreover, there could be multiple options be optimized for a given operational scenario.
to place the SDN controller and resources within the 8) Integrated SDN and NFV This area needs more inves-
MANO framework. Each option impacts the communi- tigation. Specially with practical limitations, we may not
cation to and from the SDN controller. see fully centralized and fully software based system at
least in the beginning. How these limitations affect the
realization of NFV would be important and in fact may
G. Outstanding Issues guide the SDN designs and development.
Recent years have seen massive growth in SDN related
research but there are still many open questions. The main
focus of the present paper is on the major challenges. In V. R ELATED W ORK
this section, we limit ourselves to the open issues and Although, the development in the area of SDN-based mobile
problems in architectural domain only. Summarizing the lit- networks is trailing behind the work in wired and fixed
erature in the current section, we identified the following networks but there are still a number of survey papers avail-
important issues which should be looked at by the research able in this domain. Now the question is, why do we need one
community. more? In this section, we present a brief summary of the avail-
1) Policy compression PCRF policies and rules need a able surveys and while they all target to explore some aspect of
consistent aggregation mechanism to be able to trans- the complex problem and moves forward the collective under-
late into routing rule for commercially available and standing of the research community, none has focused on the
OpenFlow-enabled routers and switches. issues delaying the realization of SDN-based mobile systems.
2) Access network layout CloudRAN seems to be the As per our knowledge, this is the first attempt to provide a
choice for SDN-based access network but the fronthaul holistic picture of the state of the art with emphasis on the
capacity and latency requirements create big hurdles weakest links.
in the production deployment. 3GPP’s LTE Release 12 The survey in [177], although is relatively recent but,
also defines Phantom cells to realize the densification of presents a case for SDN by highlighting the issues with the
base-stations. Thorough investigation is needed to find current networks, such as, heterogeneous access networks,
the optimal alternative for different operating scenarios. network ossification, increase in cost for operator amid dwin-
3) Signaling/data split Separation of control signaling and dling revenues, etc., and how SDN can provide solutions. The
data transmission is proposed to achieve huge energy paper, then, summarizes work in SDN and NFV for mobile
efficiency in BCG2 architecture and Phantom cell con- and wireless networks and identifies open technical issues,
cept (see Section II-C). We found only three papers such as, heterogeneous system’s support, decomposition of
which proposed to integrate the idea into a SDN-based control functions from protocol layers, scalability, customized
mobile network architecture [11]. The 85-90% saving and open API development, virtual machine migration, etc.
potential, as reported in [21], is significant enough to The paper also advocates the joint design of SDN and NFV.
encourage more exploration in this area. While design- SDN provides unprecedented visibility into the network allow-
ing the architecture for a new system, it may not be ing virtualization functions to have a clean abstraction to slice
difficult to adopt the signaling/data split paradigm. the resources.
4) Level of centralization Moving all functionalities to The survey in [178] discusses all SDN related research
cloud may not be a desirable choice for most of the in wireless domain, including wireless LAN, Cellular,
operating scenarios. One reason is the limitations posed mesh networks, sensor networks, etc. Three publications-
by GPP in processing real-time signals. Another is the SoftCell [66], CellSDN [4], and SoftRAN [3]- related to cel-
inhibitive cost of fiber deployment if it is not already lular networks are mentioned in the survey, not surprising for
available. This area is being explored, mostly in terms such an early survey paper. The paper mostly lists the publi-
of cost, but comprehensive investigation is still missing cations for each class of wireless networks and identifies the
which can also look at the partial centralization scenarios respective advantages and technical issues, such as, latency
and quantify the negatives, such as, the inability to carry and channel variability for power allocation.
out CoMP etc. An almost similar list of publications in cellular domain
5) Scalability Inter-cluster communication between the appeared in [179], i.e., SoftCell [66], SoftRAN [3],
controllers responsible for clusters of base-stations has Mobileflow [63], and OpenRadio [6]. This paper discusses
not been explored at all. the benefits, SDN brings to the access networks, such as,
ZAIDI et al.: WILL SDN BE PART OF 5G? 3251

virtualization, interference management, and mobile traffic Another recent survey [184] briefly surveys all technologies
management. The challenges and issues in realization of SDN- and applications associated with 5G. The survey also touches
based mobile networks are also briefly discussed, such as, upon SDN and only superficially covers some research work
actual cost saving margins, scalability of controller, secu- under the theme. A more in-depth analysis of some of SDN-
rity, isolation between network slices, estimation of channel based mobile network architectures, i.e., [3], [4], [6], [8], [63],
load for proper resource allocation, and handover between [66], [136], [153], and [174] are presented in [185] in terms
different service providers. Similar survey of SoftCell [66], of ideas presented in the proposals and their limitations. The
SoftRAN [3], Mobileflow [63], and CROWD project [136] survey in [186] looks at the proposals for softwarization and
can be found in [180]. The paper also discussed open research cloudification of cellular networks in terms of optimization
problems such as SDN-enabled cross-layer MIMO and hetero- and provisions for energy harvesting for sustainable future.
geneous radio technologies. The gaps in the technologies are also identified. All of the
The survey in [181] has a brief section on architectural above mentioned surveys, however, have a broader scope than
designs of SDN-based mobile networks and they have classi- just SDN-based mobile network architecture and they have
fied them according to their closeness to CloudRAN or mobile only looked at some SDN papers appropriate for the major
edge computing. This paper is rather an overview of dif- theme of their survey papers.
ferent issues including a brief history, high-level discussion
about business case and challenges and technical problems, VI. S UMMARY
and standardization activities in this space. In this paper, we argue that transforming the current mobile
In a recent and most comprehensive survey of SDN and vir- network infrastructure to a SDN-enabled architecture may take
tualization research for LTE mobile networks [72], the authors more time than expected. With 5G around the corner (5G NR
have provided a general overview of SDN and virtualization (New Radio) non-standalone standard approved in December
technologies and their respective benefits. They have devel- 2017 and full standard is expected in 2018), it is likely that we
oped a taxonomy to survey the research space based on the may just see some point solutions based on SDN rather than
elements of modern cellular systems, e.g., access network, core a full scale roll-out of the technology. We identified six major
network, and backhaul. Within each class, the author further roadblocks in realization of SDN-based mobile networks, i.e.,
classified the material in terms of relevant topics, such as, fronthaul, latency of general purpose platforms, backward
resource virtualization, resource abstraction, mobility manage- compatibility, disruptive deployment, SDN specific security
ment, etc. They have also looked at the use cases in each class. vulnerabilities, and clear and compelling business case. We
It is the most comprehensive survey one could find in the also looked at the state of the solutions under each issue. The
radio access network research relevant to SDN. The thrust of major lessons learned from our study are summarized in the
the survey is complementary to the present paper. If the read- following subsections.
ers want better understanding about the material covered in
Section IV, they are recommended to read [72]. On the other A. Fronthaul
hand, the open challenges briefly discussed at the end of [72]
and the relevant work under each challenge are discussed in 1) Fronthaul’s capacity and latency requirements need opti-
detail in the present paper. cal fiber links between the remote radio units (RRU)
The survey in [182] covers the holistic wireless space and the centralized processing unit. This is perhaps the
including cellular, WLAN, mesh, sensor, and home networks. biggest hurdle facing SDN deployments in places where
In cellular networks, the major contribution of this paper is fiber is not available. Countries with rich fiber coverage
to cover only the significant papers in RAN and core network are in an advantageous position.
architectures with the focus on the inclusion of virtualization 2) Microwave can be used in place of optical fiber but only
in the design and use of OpenFlow. The present paper pro- for low rate CPRI (upto 2.5Gbits/s, future advances may
vides much deeper discussion about the work in cellular SDN support up to 10Gbps).42
space and a much comprehensive coverage of the research 3) CPRI, an internal base-station protocol to transport dig-
area. itized waveform from cabinet to antenna heads, is the
A much recent survey [183] covers the proposals for most widely used transport protocol for RRU-cloud con-
SDN/NFV architectures for EPC, i.e., core network only. The nection but it has the worst data transport efficiency
authors found three different approaches to re-architect EPC, e.g., a 20MHz LTE channel can carry up to 150Mbps
i.e., virtualizing EPC (vEPC) using NFV, decoupling con- in downlink but requires CPRI rate of 2.5Gbps [29].
trol and user planes in vEPC with SDN technology, and 4) Other than capacity, delay and jitter are important issues
fully SDN realized core network. They then classified the for fronthaul. According to [31], delay should remain
proposals according to the four attributes, namely, 1- architec- under 100 microsecond and jitter should be under 65
tural approach (evolutionary or revolutionary), 2-technology nanoseconds for CPRI.
adoption (full or partial adoption of SDN or NFV or inte- 5) Functional split is a promising solution to cater for fron-
grated NFV/SDN technology), 3- functional implementation thaul capacity constraints. It is highly probable that layer
(migration to VMs, decomposition of functions, merging of 1 and part of layer 2 would remain at the remote antenna
multiple functions into single unit), and 4- deployment strategy 42 http://www.cablefree.net/wirelesstechnology/4glte/cpri-front-haul-
(distributed or centralized). technology/
3252 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

sites, i.e., at RRUs and the rest of the layered protocol E. SDN Specific Security Issues
would move to the cloud. There is no clear winner in this Security is perhaps the most researched area in SDN
case. You loose multiplexing gain when move away from domain. Although, the development in wired networks are
full centralization option. Operating conditions and fron- more advanced than in the wireless networks. SDN brings
thaul constraints are the main drivers to choose the split additional vulnerabilities to the network but at the same time
option. Chinamobile’s NGFI [35] is also a promising provide enormous ease in controlling the network as well as
option in practical deployments. visibility throughout the system. Exploiting these properties,
6) Use of Ethernet in fronthaul with or without CPRI is better security systems can be introduced and implemented.
also very likely in future as it is a mature transport The salient points from our study are given below.
technology with much better efficiency than CPRI. 1) Controller is the most critical component of the SDN
7) The most important development in this regard is IEEE network and its compromise would give tremendous
1914 working group (NGFI). The working group has access to the attacker. DoS or DDoS attack on the con-
two active projects: IEEE 1914.1 is studying NGFI of troller would also be extremely devastating. Distributed
Chinamobile and IEEE 1914.3 is looking at encapsula- control plane is an important paradigm for reliable
tion of digitized radio signal (I/Q samples) into Ethernet and scalable network control which should be explored
frames for fronthaul transmission. in detail and the inter-controller interfaces should be
defined.
B. Latency of GPP 2) There is still lack of standardization for secure north-
GPPs are not designed for real-time signal processing and bound and southbound interfaces. A dedicated link can
the delays could be of the order of tens of microseconds [48]. connect controller to its switches but appropriate level
There has been significant efforts in implementing the base- of encryption is needed to provide secure communica-
station components in GPP. So far, the data rates up to tion. The monetary implications of dedicated link and
43.8Mbps are supported on a 20MHz channel by an imple- feasibility should be studied.
mentation of LTE uplink Rx PHY in GPP by Microsoft [48]. 3) Programmability of the system may lead to policy incon-
It is still a long way to go as far as 5G goals of 1000x capacity sistencies and configuration errors. There are real-time
and 100x data rate are concerned. tools for policy control resolution but they have scalabil-
Hybrid designs with some hardware accelerators for highly- ity issues while dealing with large applications or large
computational blocks, e.g., turbo coding may provide a prac- networks.
tical approach to meet the targets. GPP latency is shown to 4) Independent experimental evaluations and comprehen-
have decreased with processor speed and number of processor sive assessment of solutions are required to check them
cores. for robustness, interoperability with third-party applica-
tions, integration of virtualization, and fast and dynamic
C. Backward Compatibility reconfiguration.
5) GPPs should also be hardened so the weaknesses in the
A practical realization of SDN should be able to work with
processors cannot be exploited to gain control over the
the legacy systems specially 4G systems. The major property
network or damage it. Moreover the current controller
to hold is to keep the LTE interfaces and GTP tunneling in the
implementations, i.e., Floodlight, OpenDayLight, POX,
new system as well. From innovation point of view, it is not
and Beacon, also have resilience issues and common
a good idea as with time better alternatives are also surfacing
application bugs are enough to crash them.
e.g., MPLS labeling provides a faster alternative than GTP tun-
6) Real-time tools are needed to verify complete logical
neling. Moreover, there has not been any work on UE handover
isolation between the virtual network sharing the same
between the two systems and regarding their co-existence in
physical resources.
the same geographical area. The recently approved 5G NR
7) It is shown in [80], that there are no solution till date
non-standalone standard defines additional interfaces and an
for data leakage and modification in SDN networks.
evolved LTE eNodeB, i.e., eLTE eNB, to provide backward
However, encryption is sought to be a valid solution
compatibility to 4G systems [2].
for SDN networks as in traditional networks but the
physical separation between controller and switches
D. Disruptive Deployment requires very important control information to be pass
An evolutionary approach for deployment is considered bet- over the communication links. Their security should
ter and acceptable to the network owners and operators as any be comprehensively tested for the available encryption
disruption would cause huge revenue loss. The ideal deploy- standards.
ment trajectory will consist of point solutions leading to the 8) Most of the development is focused on OpenFlow
full SDN-based systems. As seen before, the LTE interfaces and should be extended for broader SDN deployments.
and GTP tunneling are necessary for interworking with legacy Moreover, OpenFlow cannot handle switch mobility yet.
systems and are kept in all proposals for evolutionary deploy- 9) Security research related to SDN-based mobile networks
ments studied in this paper. Most of the ideas, however, still is lagging behind. Wireless shared medium, mobility
remains at a conceptual level and prototype implementation between different access technologies and different
are necessary to ascertain their workability. operators’ network, constrained computational capacity,
ZAIDI et al.: WILL SDN BE PART OF 5G? 3253

and backward compatibility to legacy networks are [2] System Architecture for the 5G System, Release 15, 3GPP Standard
termed as important security issues relevant only to TS 23.501 v15.0.0 (2017.12), Dec. 2017. [Online]. Available:
http://www.3gpp.org/DynaReport/23501.htm
mobile networks. [3] A. Gudipati, D. Perry, L. E. Li, and S. Katti, “SoftRAN:
Software defined radio access network,” in Proc. 2nd ACM
SIGCOMM Workshop Hot Topics Softw. Defined Netw.,
F. Clear and Compelling Business Case New York, NY, USA: ACM, 2013, pp. 25–30. [Online]. Available:
Virtualization, use of GPP instead of specialized hardware, http://doi.acm.org/10.1145/2491185.2491207
[4] L. E. Li, Z. M. Mao, and J. Rexford, “Toward software-defined cellular
and resource pooling in the cloud are the key ideas behind networks,” in Proc. Eur. Workshop Softw. Defined Netw. (EWSDN),
CAPEX and OPEX savings in SDN networks. Actual savings Oct. 2012, pp. 7–12.
depend on the operating conditions, such as, availability or [5] S. Bhaumik et al., “CloudiQ: A framework for processing base stations
non-availability of optical fiber fronthaul, traffic load, equip- in a data center,” in Proc. 18th Annu. Int. Conf. Mobile Comput. Netw.,
New York, NY, USA: ACM, 2012, pp. 125–136. [Online]. Available:
ment cost etc. In some business studies, 68% lower CAPEX http://doi.acm.org/10.1145/2348543.2348561
and 67% lower OPEX are reported when LTE EPC is virtu- [6] M. Bansal, J. Mehlman, S. Katti, and P. Levis, “OpenRadio: A pro-
alized. Although, virtualization is not dependent on SDN and grammable wireless dataplane,” in Proc. 1st Workshop Hot Topics
Softw. Defined Netw., New York, NY, USA: ACM, 2012, pp. 109–114.
immediate gains are possible when traditional methods are [Online]. Available: http://doi.acm.org/10.1145/2342441.2342464
used for virtualization. SDN, however, provides a sustainable [7] M. Yang et al., “OpenRan: A software-defined ran archi-
business model for long-term future. tecture via virtualization,” SIGCOMM Comput. Commun. Rev.,
vol. 43, no. 4, pp. 549–550, Aug. 2013. [Online]. Available:
Some analyst also believe that the cost reduction due to http://doi.acm.org/10.1145/2534169.2491732
GPP is based on the assumption that specialized hardware is [8] I. F. Akyildiz, P. Wang, and S.-C. Lin, “SoftAir: A soft-
over-priced which is not always true. Considering the cost of ware defined networking architecture for 5G wireless systems,”
transforming telco companies to suit the new SDN paradigm, Comput. Netw., vol. 85, pp. 1–18, Jul. 2015. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1389128615001632
the cost reduction may not be as great as predicted in some [9] J. Lai, L. Jiang, M. Lei, A. Abdollahpouri, and W. Fang, “Software-
studies. defined cellular networking: A practical path towards 5G,” Int. J.
However, this has also been pointed out clearly that the real Commun. Netw. Distrib. Syst., vol. 14, no. 1, pp. 89–105, Nov. 2015,
doi: 10.1504/IJCNDS.2015.066019.
benefit of SDN is not in the cost savings but it is in provid- [10] S. M. Raza, D. S. Kim, and H. Choo, “The proposal for SDN supported
ing the fail-fast capability to the networks. Carriers with such future 5G networks,” in Proc. Conf. Res. Adapt. Convergent Syst.,
capability, where the deployment cycle for new technologies New York, NY, USA: ACM, 2014, pp. 180–185. [Online]. Available:
http://doi.acm.org/10.1145/2663761.2664237
can be cut down to hours from days, will have a unique edge
[11] J. Liu, T. Zhao, S. Zhou, Y. Cheng, and Z. Niu, “CONCERT: A cloud-
over others and eventually nobody could survive without it. based architecture for next-generation cellular systems,” IEEE Wireless
The disruptive model of transformation of communication Commun., vol. 21, no. 6, pp. 14–22, Dec. 2014.
network with full softwarization and centralization is very [12] J. Wu, S. Guo, J. Li, and D. Zeng, “Big data meet green chal-
lenges: Greening big data,” IEEE Syst. J., vol. 10, no. 3, pp. 873–887,
unlikely to be followed as none of the saving promises and Sep. 2016.
monetary benefits can justify them. An incremental approach [13] L. Chiaraviglio et al., “Bringing 5G into rural and low-income areas:
where SDN is introduced through point solutions in an already Is it feasible?” IEEE Commun. Stand. Mag., vol. 1, no. 3, pp. 50–57,
Sep. 2017.
erected network seems to be more plausible. [14] M. Agiwal, A. Roy, and N. Saxena, “Next generation 5G wireless
networks: A comprehensive survey,” IEEE Commun. Surveys Tuts.,
G. Current State of the Art vol. 18, no. 3, pp. 1617–1655, 3rd Quart., 2016.
[15] Z. Zaidi, V. Friderikos, and M. A. Imran, “Future RAN architecture:
In our summary of architectural developments in SDN- SD-RAN through a general-purpose processing platform,” IEEE Veh.
based mobile networks, we list the repeated themes in the Technol. Mag., vol. 10, no. 1, pp. 52–60, Mar. 2015.
[16] S. Abdelwahab, B. Hamdaoui, M. Guizani, and T. Znati, “Network
literature, such as, EPC nodes are mostly featured as applica- function virtualization in 5G,” IEEE Commun. Mag., vol. 54, no. 4,
tion plane modules, SDN-based mobile networks may be close pp. 84–91, Apr. 2016.
to CloudRAN in design with partial centralization where full [17] Network Function Virtualisation (NFV); Management and
Orchestration, ETSI NFV ISG Standard GS NFV-MAN 001
centralization is costly, distributed control plane is scalable, V1.1.1, Dec. 2014.
OpenFlow is preferred choice of researchers at least for now, [18] “The LTE network architecture: A comprehensive tutorial,” in LTE—
and an integrated SDN/NFV design is more desirable from The UMTS Long Term Evolution: From Theory to Practice. Chichester,
business point of view. Outstanding issues are discussed in U.K.: Wiley, 2009. [Online]. Available: www.alcatel-lucent.com/4g-
consumer-communications
Section IV-G. [19] G. Rittenhouse, “Green wireless networks,” presented at the Keynote
There is still a gap between the state of the art and the IEEE WCNC 2012, Apr. 2010, pp. 1–19. [Online]. Available:
point where SDN would be tech-ready for production deploy- http://wcnc2012.ieee-wcnc.org/Keynote.html
[20] H. Ishii, Y. Kishiyama, and H. Takahashi, “A novel architecture for
ment. SDN provides a novel and challenging playing field for LTE-B: C-plane/U-plane split and phantom cell concept,” in Proc. IEEE
research community but it is important to get the priorities Globecom Workshops (GC Wkshps), Anaheim, CA, USA, Dec. 2012,
right and emphasis should be on solving the problems which pp. 624–630.
matter the most. [21] I. Godor, “D3.3: Final report on green network technolo-
gies,” Alcatel-Lucent Deutschland AG, Stuttgart, Germany, Rep.
EARTH INFSO-ICT-247733, Jun. 2012. [Online]. Available:
R EFERENCES https://www.ictearth.eu/publications/deliverables/deliverables.html
[22] “Cloud RAN: Benefits of centralization an virtualization,”
[1] C. Gallon, “Carrier software defined networking,” Ofcom, Santa Clara, CA, USA, Aricent, White Paper, 2016. [Online].
London, U.K., Rep., Mar. 2014. [Online]. Available: https:// Available: https://www.aricent.com/whitepapers/cloud-ran-benefits-
www.ofcom.org.uk/__data/assets/pdf_file/0013/32143/sdn_report.pdf centralization-and-virtualization
3254 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

[23] C. Ge, Z. Sun, N. Wang, K. Xu, and J. Wu, “Energy management [45] A. S. Thyagaturu, Y. Dashti, and M. Reisslein, “SDN-based smart
in cross-domain content delivery networks: A theoretical perspec- gateways (Sm-GWs) for multi-operator small cell network manage-
tive,” IEEE Trans. Netw. Service Manag., vol. 11, no. 3, pp. 264–277, ment,” IEEE Trans. Netw. Service Manag., vol. 13, no. 4, pp. 740–753,
Sep. 2014. Dec. 2016.
[24] K. Yang, S. Martin, C. Xing, J. Wu, and R. Fan, “Energy-efficient [46] M. Artuso, A. Marcano, and H. Christiansen, “Cloudification of
power control for device-to-device communications,” IEEE J. Sel. mmwave-based and packet-based fronthaul for future heterogeneous
Areas Commun., vol. 34, no. 12, pp. 3208–3220, Dec. 2016. mobile networks,” IEEE Wireless Commun., vol. 22, no. 5, pp. 76–82,
[25] K. Wang, Y. Wang, Y. Sun, S. Guo, and J. Wu, “Green industrial Oct. 2015.
Internet of Things architecture: An energy-efficient perspective,” IEEE [47] A. S. Thyagaturu, A. Mercian, M. P. McGarry, M. Reisslein, and
Commun. Mag., vol. 54, no. 12, pp. 48–54, Dec. 2016. W. Kellerer, “Software defined optical networks (SDONs): A com-
[26] J. An et al., “Achieving sustainable ultra-dense heterogeneous networks prehensive survey,” IEEE Commun. Surveys Tuts., vol. 18, no. 4,
for 5G,” IEEE Commun. Mag., vol. 55, no. 12, pp. 84–90, Dec. 2017. pp. 2738–2786, 4th Quart., 2016.
[27] J. Wu, “Green wireless communications: From concept to reality [48] K. Tan et al., “Sora: High-performance software radio using general-
[industry perspectives],” IEEE Wireless Commun., vol. 19, no. 4, purpose multi-core processors,” Commun. ACM, vol. 54, no. 1,
pp. 4–5, Aug. 2012. pp. 99–107, Jan. 2011. [Online]. Available: http://doi.acm.org/10.1145/
[28] Q. Luo, W. Fang, J. Wu, and Q. Chen, “Reliable broadband wireless 1866739.1866760
communication for high speed trains using baseband cloud,” EURASIP [49] N. Kai, S. Jianxing, C. Kuilin, and K. K. Chai, “TD-LTE
J. Wireless Commun. Netw., vol. 2012, no. 1, p. 285, Sep. 2012. eNodeB prototype using general purpose processor,” in Proc. 7th
[Online]. Available: https://doi.org/10.1186/1687-1499-2012-285 Int. ICST Conf. Commun. Netw. China (CHINACOM), Aug. 2012,
[29] K. Murphy, “Centralized ran and fronthaul,” Stockholm, Sweden, pp. 822–827.
Ericsson Inc., White Paper, May 2015. [Online]. Available: [50] Z. Zhu et al., “Virtual base station pool: Towards a wireless network
http://www.ospmag.com/files/pdf/whitepaper/C-RAN_and_Fronthaul_ cloud for radio access networks,” in Proc. 8th ACM Int. Conf.
White_Paper.pdf Comput. Front., New York, NY, USA: ACM, 2011, pp. 1–10. [Online].
[30] M. Peng, C. Wang, V. Lau, and H. V. Poor, “Fronthaul-constrained Available: http://doi.acm.org/10.1145/2016604.2016646
cloud radio access networks: Insights and challenges,” IEEE Wireless [51] C. Y. Yeoh, M. H. Mokhtar, A. A. A. Rahman, and A. K. Samingan,
Commun., vol. 22, no. 2, pp. 152–160, Apr. 2015. “Performance study of LTE experimental testbed using
[31] D. Chitimalla, K. Kondepu, L. Valcarenghi, M. Tornatore, and OpenAirInterface,” in Proc. 18th Int. Conf. Adv. Commun.
B. Mukherjee, “5G fronthaul-latency and jitter studies of CPRI over Technol. (ICACT), Jan. 2016, pp. 617–622.
Ethernet,” IEEE J. Opt. Commun. Netw., vol. 9, no. 2, pp. 172–182, [52] J. Wang, J. Xu, Y. Yang, and H. Xu, “GPP based open cellular network
Feb. 2017. towards 5G,” China Commun., vol. 14, no. 6, pp. 189–198, 2017.
[32] N. J. Gomes, P. Chanclou, P. Turnbull, A. Magee, and V. Jungnickel, [53] M. Moradi, Y. Lin, Z. M. Mao, S. Sen, and O. Spatscheck, “SoftBox:
“Fronthaul evolution: From CPRI to Ethernet,” Opt. Fiber A customizable, low-latency, and scalable 5G core network archi-
Technol., vol. 26, pp. 50–58, Dec. 2015. [Online]. Available: http:// tecture,” IEEE J. Sel. Areas Commun., vol. 36, no. 3, pp. 438–456,
www.sciencedirect.com/science/article/pii/S1068520015000942 Mar. 2018.
[54] I. Gomez-Miguelez et al., “srsLTE: An open-source platform
[33] M. Peng, Y. Li, J. Jiang, J. Li, and C. Wang, “Heterogeneous cloud
for LTE evolution and experimentation,” in Proc. 10th ACM Int.
radio access networks: A new perspective for enhancing spectral
Workshop Wireless Netw. Testbeds Exp. Eval. Characterization
and energy efficiencies,” IEEE Wireless Commun., vol. 21, no. 6,
(WiNTECH), New York, NY, USA: ACM, 2016, pp. 25–32.
pp. 126–135, Dec. 2014.
[Online]. Available: http://doi.acm.org/10.1145/2980159.2980163,
[34] “OpenFog architecture overview,” Fremont, CA, USA, OpenFog
doi: 10.1145/2980159.2980163.
Consortium, White Paper, Feb. 2016. [Online]. Available: https://
[55] W. Rouwet, “Next generation access network development platform
www.openfogconsortium.org/wp-content/uploads/OpenFog-
and vAccess,” Tempe, AZ, USA, NXP Semicond., White Paper,
Architecture-Overview-WP-2-2016.pdf
Mar. 2015. [Online]. Available: www.nxp.com
[35] C.-l. I et al., “Rethink fronthaul for soft RAN,” IEEE Commun. Mag., [56] T. Kazaz, C. Van Praet, M. Kulin, P. Willemen, and I. Moerman,
vol. 53, no. 9, pp. 82–88, Sep. 2015. “Hardware accelerated SDR platform for adaptive air interfaces,” in
[36] N. Nikaein, “Processing radio access network functions in the cloud: Proc. ETSI Workshop Future Radio Technol. Air Interfaces, 2016,
Critical issues and modeling,” in Proc. 6th Int. Workshop Mobile pp. 1–10.
Cloud Comput. Services, New York, NY, USA: ACM, 2015, pp. 36–43. [57] 3GPP, “study on new radio access technology: Radio access
[Online]. Available: http://doi.acm.org/10.1145/2802130.2802136 architecture and interfaces, release 14, v14.0.0 (2017.03),” 3GPP,
[37] A. S. Thyagaturu, Z. Alharbi, and M. Reisslein, “R-FFT: Function Washington, DC, USA, Rep. TR 38.801, Apr. 2017. [Online].
split at IFFT/FFT in unified LTE CRAN and cable access network,” Available: http://www.3gpp.org/DynaReport/38801.htm
CoRR, vol. abs/1708.08902, pp. 1–15, Aug. 2017. [Online]. Available: [58] N. McKeown et al., “OpenFlow: Enabling innovation in cam-
http://arxiv.org/abs/1708.08902 pus networks,” SIGCOMM Comput. Commun. Rev., vol. 38,
[38] G. Mountaser, M. L. Rosas, T. Mahmoodi, and M. Dohler, “On the fea- no. 2, pp. 69–74, Mar. 2008. [Online]. Available: http://doi.acm.org/
sibility of MAC and PHY split in cloud RAN,” in Proc. IEEE Wireless 10.1145/1355734.1355746
Commun. Netw. Conf., IEEE, Mar. 2017, pp. 1–6. [59] D. Levin, M. Canini, S. Schmid, F. Schaffert, and A. Feldmann,
[39] A. Checko, A. P. Avramova, M. S. Berger, and H. L. Christiansen, “Panopticon: Reaping the benefits of incremental SDN deploy-
“Evaluating C-RAN fronthaul functional splits in terms of network ment in enterprise networks,” in Proc. USENIX Annu. Tech.
level energy and cost savings,” J. Commun. Netw., vol. 18, no. 2, Conf. (USENIX ATC), Philadelphia, PA, USA: USENIX Assoc.,
pp. 162–172, Apr. 2016. Jun. 2014, pp. 333–345. [Online]. Available: https://www.usenix.org/
[40] C.-Y. Chang, R. Schiavi, N. Nikaein, T. Spyropoulos, and C. Bonnet, conference/atc14/technical-sessions/presentation/levin
“Impact of packetization and functional split on c-ran fronthaul [60] M. Markovitch and S. Schmid, “SHEAR: A highly available and flexi-
performance,” in Proc. IEEE Int. Conf. Commun. (ICC), May 2016, ble network architecture marrying distributed and logically centralized
pp. 1–7. control planes,” in Proc. IEEE 23rd Int. Conf. Netw. Protocols (ICNP),
[41] C.-Y. Chang, N. Nikaein, and T. Spyropoulos, “Impact of packetization San Francisco, CA, USA, Nov. 2015, pp. 78–89.
and scheduling on C-RAN fronthaul performance,” in Proc. IEEE Glob. [61] C. Jin et al., “Exerting fine-grained path control over legacy
Commun. Conf. (GLOBECOM), Dec. 2016, pp. 1–7. switches in hybrid networks,” Univ. Minnesote, Minneapolis,
[42] A. D. L. Oliva et al., “Xhaul: Toward an integrated fronthaul/backhaul MN, USA, Rep. TR 16-035, Sep. 2016. [Online]. Available:
architecture in 5G networks,” IEEE Wireless Commun., vol. 22, no. 5, https://www.cs.umn.edu/sites/cs.umn.edu/files/tech_reports/16-035.pdf
pp. 32–40, Oct. 2015. [62] C. Jin, C. Lumezanu, Q. Xu, Z.-L. Zhang, and G. Jiang, “Telekinesis:
[43] G. Edwards et al., Fronthaul Evolution Toward 5G: Standards and Controlling legacy switch routing with OpenFlow in hybrid networks,”
Proof of Concepts, Design Reuse, Grenoble, France, 2016. [Online]. in Proc. 1st ACM SIGCOMM Symp. Softw. Defined Netw. Res.,
Available: https://www.design-reuse.com/articles/40008/fronthaul- New York, NY, USA: ACM, 2015, pp. 1–7. [Online]. Available:
evolution-toward-5g-standards-and-proof-of-concepts.html http://doi.acm.org/10.1145/2774993.2775013
[44] M. Z. Shakir et al., “Smart backhauling and fronthauling for 5G [63] K. Pentikousis, Y. Wang, and W. Hu, “Mobileflow: Toward software-
networks: From precoding to network architecture [guest editorial],” defined mobile networks,” IEEE Commun. Mag., vol. 51, no. 7,
IEEE Wireless Commun., vol. 22, no. 5, pp. 10–12, Oct. 2015. pp. 44–53, Jul. 2013.
ZAIDI et al.: WILL SDN BE PART OF 5G? 3255

[64] J. Kempf, B. Johansson, S. Pettersson, H. Lüning, and T. Nilsson, [85] D. Kreutz et al., “Software-defined networking: A compre-
“Moving the mobile evolved packet core to the cloud,” in Proc. IEEE hensive survey,” Proc. IEEE, vol. 103, no. 1, pp. 14–76,
8th Int. Conf. Wireless Mobile Comput. Netw. Commun. (WiMob), Jan. 2015.
Barcelona, Spain, Oct. 2012, pp. 784–791. [86] T. Dargahi, A. Caponi, M. Ambrosin, G. Bianchi, and M. Conti, “A
[65] N. Cvijetic, A. Tanaka, K. Kanonakis, and T. Wang, “SDN-controlled survey on the security of stateful SDN data planes,” IEEE Commun.
topology-reconfigurable optical mobile fronthaul architecture for bidi- Surveys Tuts., vol. 19, no. 3, pp. 1701–1725, 3rd Quart., 2017.
rectional comp and low latency inter-cell D2D in the 5G mobile [87] V. Vassilakis, I. Moscholios, B. Alzahrani, and M. Logothetis, “On
era,” Opt. Exp., vol. 22, no. 17, pp. 20809–20815, Aug. 2014. the security of software-defined next-generation cellular networks,” in
[Online]. Available: http://www.opticsexpress.org/abstract.cfm?URI= Proc. IEICE Inf. Commun. Technol. Forum (ICTF), Jul. 2016, pp. 1–6.
oe-22-17-20809 [Online]. Available: http://repository.uwl.ac.uk/2827/
[66] X. Jin, L. E. Li, L. Vanbever, and J. Rexford, “SoftCell: [88] G. Hampel, M. Steiner, and T. Bu, “Applying software-defined
Scalable and flexible cellular core network architecture,” in networking to the telecom domain,” in Proc. IEEE Conf. Comput.
Proc. 9th ACM Conf. Emerg. Netw. Exp. Technol. (CoNEXT), Commun. Workshops (INFOCOM WKSHPS), Turin, Italy, Apr. 2013,
New York, NY, USA: ACM, 2013, pp. 163–174. [Online]. Available: pp. 133–138.
http://doi.acm.org/10.1145/2535372.2535377 [89] S. Namal, I. Ahmad, A. Gurtov, and M. Ylianttila, “Enabling secure
[67] P. Demestichas et al., “5G on the horizon: Key challenges for the radio- mobility with OpenFlow,” in Proc. IEEE SDN Future Netw. Services
access network,” IEEE Veh. Technol. Mag., vol. 8, no. 3, pp. 47–53, (SDN4FNS), Trento, Italy, Nov. 2013, pp. 1–5.
Sep. 2013. [90] R. Guerzoni, R. Trivisonno, and D. Soldani, “SDN-based architecture
[68] R. Trivisonno, R. Guerzoni, I. Vaishnavi, and D. Soldani, “SDN-based and procedures for 5G networks,” in Proc. 1st Int. Conf. 5G Ubiquitous
5G mobile networks: Architecture, functions, procedures and backward Connectivity, Nov. 2014, pp. 209–214.
compatibility,” Trans. Emerg. Telecommun. Technol., vol. 26, no. 1, [91] R. Atat et al., “Enabling cyber-physical communication in 5G cellular
pp. 82–92, Jan. 2015, doi: 10.1002/ett.2915. networks: Challenges, spatial spectrum sensing, and cyber-security,”
[69] C. J. Bernardos et al., “An architecture for software defined wire- IET Cyber Phys. Syst. Theory Appl., vol. 2, no. 1, pp. 49–54,
less networking,” IEEE Wireless Commun., vol. 21, no. 3, pp. 52–61, Apr. 2017.
Jun. 2014. [92] Z. Yan, P. Zhang, and A. V. Vasilakos, “A security and trust framework
[70] Y. Kyung, T. M. Nguyen, K. Hong, J. Park, and J. Park, “Software for virtualized networks and software-defined networking,” Security
defined service migration through legacy service integration into 4G Commun. Netw., vol. 9, no. 16, pp. 3059–3069, 2016, sCN-14-0760.R1.
networks and future evolutions,” IEEE Commun. Mag., vol. 53, no. 9, [Online]. Available: http://dx.doi.org/10.1002/sec.1243
pp. 108–114, Sep. 2015. [93] X. Duan and X. Wang, “Authentication handover and privacy protec-
[71] J. Costa-Requena et al., “SDN and NFV integration in gener- tion in 5G hetnets using software-defined networking,” IEEE Commun.
alized mobile network architecture,” in Proc. Eur. Conf. Netw. Mag., vol. 53, no. 4, pp. 28–35, Apr. 2015.
Commun. (EuCNC), Paris, France, Jun. 2015, pp. 154–158.
[94] Y. Fu, Z. Yan, H. Li, X. L. Xin, and J. Cao, “A secure SDN
[72] V.-G. Nguyen, T.-X. Do, and Y. Kim, “SDN and virtualization-
based multi-RANs architecture for future 5G networks,” Comput.
based LTE mobile network architectures: A comprehensive survey,”
Security, vol. 70, pp. 648–662, Sep. 2017. [Online]. Available:
Wireless Pers. Commun., vol. 86, no. 3, pp. 1401–1438, 2016,
http://www.sciencedirect.com/science/article/pii/S0167404817301785
doi: 10.1007/s11277-015-2997-7.
[95] S. Luo, M. Dong, K. Ota, J. Wu, and J. Li, “A security assessment
[73] P. Rost et al., “Cloud technologies for flexible 5G radio access
mechanism for software-defined networking-based mobile networks,”
networks,” IEEE Commun. Mag., vol. 52, no. 5, pp. 68–76, May 2014.
Sensors, vol. 15, no. 12, pp. 31843–31858, 2015. [Online]. Available:
[74] I. Ahmad, S. Namal, M. Ylianttila, and A. Gurtov, “Security in software
http://www.mdpi.com/1424-8220/15/12/29887
defined networks: A survey,” IEEE Commun. Surveys Tuts., vol. 17,
no. 4, pp. 2317–2346, 4th Quart., 2015. [96] R. Skowyra, S. Bahargam, and A. Bestavros, “Software-defined IDS
[75] M. Liyanage et al., “Security for future software defined mobile for securing embedded mobile devices,” in Proc. IEEE High Perform.
networks,” in Proc. 9th Int. Conf. Next Gener. Mobile Appl. Services Extreme Comput. Conf. (HPEC), Waltham, MA, USA, Sep. 2013,
Technol., Cambridge, U.K., Sep. 2015, pp. 256–264. pp. 1–7.
[76] A. Y. Ding, J. Crowcroft, S. Tarkoma, and H. Flinck, “Software defined [97] R. W. Skowyra, A. Lapets, A. Bestavros, and A. Kfoury, “Verifiably-
networking for security enhancement in wireless mobile networks,” safe software-defined networks for CPS,” in Proc. 2nd ACM
Comput. Netw., vol. 66, pp. 94–101, Jun. 2014. [Online]. Available: Int. Conf. High Confidence Netw. Syst. (HiCoNS), New York,
http://www.sciencedirect.com/science/article/pii/S1389128614001133 NY, USA: ACM, 2013, pp. 101–110. [Online]. Available:
[77] D. He, S. Chan, and M. Guizani, “Securing software defined wireless http://doi.acm.org/10.1145/2461446.2461461
networks,” IEEE Commun. Mag., vol. 54, no. 1, pp. 20–25, Jan. 2016. [98] R. Jin and B. Wang, “Malware detection for mobile devices using
[78] M. Dabbagh, B. Hamdaoui, M. Guizani, and A. Rayes, “Software- software-defined networking,” in Proc. 2nd GENI Res. Educ. Exp.
defined networking security: Pros and cons,” IEEE Commun. Mag., Workshop (GREE), Washington, DC, USA: IEEE Comput. Soc., 2013,
vol. 53, no. 6, pp. 73–79, Jun. 2015. pp. 81–88, doi: 10.1109/GREE.2013.24.
[79] Q. Yan, F. R. Yu, Q. Gong, and J. Li, “Software-defined networking [99] G. Hurel, R. Badonnel, A. Lahmadi, and O. Festor, Outsourcing Mobile
(SDN) and distributed denial of service (DDoS) attacks in cloud com- Security in the Cloud. Berlin, Germany: Springer, 2014, pp. 69–73,
puting environments: A survey, some research issues, and challenges,” doi: 10.1007/978-3-662-43862-6_9.
IEEE Commun. Surveys Tuts., vol. 18, no. 1, pp. 602–622, 1st Quart., [100] K. Grenier, “Business case for a common NFV platform,” ACG
2016. Res., Gilbert, AZ, USA, Rep., Aug. 2015. [Online]. Available:
[80] S. Scott-Hayward, S. Natarajan, and S. Sezer, “A survey of security http://acgcc.com/business-case-for-a-common-nfv-platform/
in software defined networks,” IEEE Commun. Surveys Tuts., vol. 18, [101] C. Bouras, P. Ntarzanos, and A. Papazois, “Cost modeling for
no. 1, pp. 623–654, 1st Quart., 2016. SDN/NFV based mobile 5G networks,” in Proc. 8th Int. Congr.
[81] C. A. Ezefibe and Y. R. Shayan, “Towards virtualisation and secured Ultra Modern Telecommun. Control Syst. Workshops (ICUMT), Lisbon,
software defined networking for wireless and cellular networks,” in Portugal, Oct. 2016, pp. 56–61.
Proc. IEEE Can. Conf. Elect. Comput. Eng. (CCECE), Vancouver, BC, [102] R. Haim, “Total cost of ownership study, virtualizing the mobile core,”
Canada, May 2016, pp. 1–5. ACG Res., Gilbert, AZ, USA, Rep., Jul. 2015. [Online]. Available:
[82] M. Liyanage, M. Ylianttila, and A. Gurtov, “Securing the control chan- http://www.affirmednetworks.com/wp-content/uploads/2015/07/TCO-
nel of software-defined mobile networks,” in Proc. IEEE Int. Symp. Report_7.13.15_ACG-Template.pdf
World Wireless Mobile Multimedia Netw., Sydney, NSW, Australia, [103] B. Naudts et al., “Techno-economic analysis of software defined
Jun. 2014, pp. 1–6. networking as architecture for the virtualization of a mobile network,”
[83] M. Chen, Y. Qian, S. Mao, W. Tang, and X. Yang, “Software- in Proc. Eur. Workshop Softw. Defined Netw., Darmstadt, Germany,
defined mobile networks security,” Mobile Netw. Appl., vol. 21, no. 5, Oct. 2012, pp. 67–72. [Online]. Available: http://sites.ieee.org/sdn4fns/
pp. 729–743, 2016, doi: 10.1007/s11036-015-0665-5. files/2013/11/SDN4FNS_panel_presentation_Bram_Naudts.pdf
[84] S. Shin, V. Yegneswaran, P. Porras, and G. Gu, “AVANT-GUARD: [104] B. Naudts, “Techno-economic analysis of software defined networking
Scalable and vigilant switch flow management in software-defined as architecture for the virtualization of a mobile network,” in Proc.
networks,” in Proc. ACM SIGSAC Conf. Comput. 38 Commun. Security IEEE Softw. Defined Netw. Future Netw. Services (SDN4FNS),
(CCS), New York, NY, USA: ACM, 2013, pp. 413–424. [Online]. Nov. 2013. [Online]. Available: http://sites.ieee.org/sdn4fns/
Available: http://doi.acm.org/10.1145/2508859.2516684 files/2013/11/SDN4FNS_panel_presentation_Bram_Naudts.pdf
3256 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

[105] A. Checko, H. L. Christiansen, and M. S. Berger, “Evaluation of energy [128] A. Bradai, K. Singh, T. Ahmed, and T. Rasheed, “Cellular software
and cost savings in mobile cloud RAN,” in Proc. OPNETWORK, defined networking: A framework,” IEEE Commun. Mag., vol. 53,
2013, pp. 1–7. [Online]. Available: http://orbit.dtu.dk/fedora/ no. 6, pp. 36–43, Jun. 2015.
objects/orbit:124961/datastreams/file_f0352259-e81d-440c-ad42- [129] A. Basta, W. Kellerer, M. Hoffmann, K. Hoffmann, and E.-D. Schmidt,
b417a02e5999/content “A virtual SDN-enabled LTE EPC architecture: A case study for S-
[106] W. Wu, L. E. Li, A. Panda, and S. Shenker, “PRAN: Programmable /P-gateways functions,” in Proc. IEEE SDN Future Netw. Services
radio access networks,” in Proc. 13th ACM Workshop Hot Topics (SDN4FNS), Trento, Italy, Nov. 2013, pp. 1–7.
Netw. (HotNets-XIII), New York, NY, USA: ACM, 2014, pp. 1–7. [130] I. Elgendi, K. S. Munasinghe, and A. Jamalipour, “A three-tier
[Online]. Available: http://doi.acm.org/10.1145/2670518.2673865 SDN architecture for DenseNets,” in Proc. 9th Int. Conf. Signal
[107] V. Suryaprakash, P. Rost, and G. Fettweis, “Are heterogeneous cloud- Process. Commun. Syst. (ICSPCS), Cairns, QLD, Australia, Dec. 2015,
based radio access networks cost effective?” IEEE J. Sel. Areas pp. 1–7.
Commun., vol. 33, no. 10, pp. 2239–2251, Oct. 2015. [131] E. B. Hamza and S. Kimura, “A scalable SDN-EPC architecture based
[108] N. Zhang and H. Hämmäinen, “Cost efficiency of SDN in LTE-based on OpenFlow-enabled switches to support inter-domain handover,”
mobile networks: Case Finland,” in Proc. Int. Conf. Workshops Netw. in Proc. 10th Int. Conf. Innov. Mobile Internet Services Ubiquitous
Syst. (NetSys), Cottbus, Germany, Mar. 2015, pp. 1–5. Comput. (IMIS), Fukuoka, Japan, Jul. 2016, pp. 272–277.
[109] A. Asensio, P. Saengudomlert, M. Ruiz, and L. Velasco, “Study of the [132] M. Yang, Y. Li, B. Li, D. Jin, and S. Chen, “Service-oriented 5G
centralization level of optical network-supported cloud RAN,” in Proc. network architecture: An end-to-end software defining approach,”
Int. Conf. Opt. Netw. Design Model. (ONDM), May 2016, pp. 1–6. Int. J. Commun. Syst., vol. 29, no. 10, pp. 1645–1657, 2016.
[110] H. Yeganeh and E. Vaezpour, “Fronthaul network design for [Online]. Available: http://www.ingentaconnect.com/content/bpl/dac/
radio access network virtualization from a CAPEX/OPEX per- 2016/00000029/00000010/art00004
spective,” Ann. Telecommun., vol. 71, no. 11, pp. 665–676, 2016, [133] F. Bannour, S. Souihi, and A. Mellouk, “Distributed SDN control:
doi: 10.1007/s12243-016-0538-3. Survey, taxonomy, and challenges,” IEEE Commun. Surveys Tuts.,
[111] M. Karimzadeh, L. Valtulina, and G. Karagiannis, “Applying vol. 20, no. 1, pp. 333–354, 1st Quart., 2018.
SDN/OpenFlow in virtualized LTE to support distributed mobility man- [134] S. Auroux, M. Dräxler, A. Morelli, and V. Mancuso, “Dynamic network
agement (DMM),” in Proc. 4th Int. Conf. Cloud Comput. Services reconfiguration in wireless DenseNets with the CROWD SDN archi-
Sci. (CLOSER), Setúbal, Portugal: SCITEPRESS, 2014, pp. 639–644, tecture,” in Proc. Eur. Conf. Netw. Commun. (EuCNC), Paris, France,
doi: 10.5220/0004946106390644. Jun./Jul. 2015, pp. 144–148.
[112] S. Kang and W. Yoon, “SDN-based resource allocation for heteroge- [135] A. Basta et al., “Towards a cost optimal design for a 5G mobile core
neous LTE and WLAN multi-radio networks,” J. Supercomput., vol. 72, network based on SDN and NFV,” IEEE Trans. Netw. Service Manag.,
no. 4, pp. 1342–1362, Apr. 2016, doi: 10.1007/s11227-016-1662-6. vol. 14, no. 4, pp. 1061–1075, Dec. 2017.
[113] S. Costanzo, D. Xenakis, N. Passas, and L. Merakos, “OpeNB: A [136] H. Ali-Ahmad et al., “An SDN-based network architecture for
framework for virtualizing base stations in LTE networks,” in Proc. extremely dense wireless networks,” in Proc. IEEE SDN Future Netw.
IEEE Int. Conf. Commun. (ICC), Sydney, NSW, Australia, Jun. 2014, Services (SDN4FNS), Trento, Italy, Nov. 2013, pp. 1–7.
pp. 3148–3153. [137] V. Yazici, U. C. Kozat, and M. O. Sunay, “A new control plane for 5G
[114] P. Spapis, K. Chatzikokolakis, N. Alonistioti, and A. Kaloxylos, “Using network architecture with a case study on unified handoff, mobility, and
SDN as a key enabler for co-primary spectrum sharing,” in Proc. 5th routing management,” IEEE Commun. Mag., vol. 52, no. 11, pp. 76–85,
Int. Conf. Inf. Intell. Syst. Appl. (IISA), Jul. 2014, pp. 366–371. Nov. 2014.
[115] D. Boviz, N. Abbas, G. Aravinthan, C. S. Chen, and M. A. Dridi, [138] R. Trivisonno, R. Guerzoni, I. Vaishnavi, and D. Soldani, “Towards
“Multi-cell coordination in cloud RAN: Architecture and optimization,” zero latency software defined 5G networks,” in Proc. IEEE Int.
in Proc. Int. Conf. Wireless Netw. Mobile Commun. (WINCOM), Fes, Conf. Commun. Workshop (ICCW), London, U.K., Jun. 2015,
Morocco, Oct. 2016, pp. 271–277. pp. 2566–2571.
[116] Y. Zhang et al., “STEERING: A software-defined networking for [139] V. G. Vassilakis, I. D. Moscholios, B. A. Alzahrani, and
inline service chaining,” in Proc. 21st IEEE Int. Conf. Netw. Protocols M. D. Logothetis, “A software-defined architecture for next-generation
(ICNP), Göttingen, Germany, Oct. 2013, pp. 1–10. cellular networks,” in Proc. IEEE Int. Conf. Commun. (ICC), Kuala
[117] K. Katsalis, N. Nikaein, E. Schiller, R. Favraud, and T. I. Braun, Lumpur, Malaysia, May 2016, pp. 1–6.
“5G architectural design patterns,” in Proc. IEEE Int. Conf. Commun. [140] X. Mi, Z. Tian, X. Xu, M. Zhao, and J. Wang, “No stack: A
Workshops (ICC), Kuala Lumpur, Malaysia, May 2016, pp. 32–37. SDN-based framework for future cellular networks,” in Proc. Int.
[118] I. Afolabi, T. Taleb, K. Samdanis, A. Ksentini, and H. Flinck, “Network Symp. Wireless Pers. Multimedia Commun. (WPMC), Sydney, NSW,
slicing & softwarization: A survey on principles, enabling technologies Australia, Sep. 2014, pp. 497–502.
& solutions,” IEEE Commun. Surveys Tuts., to be published. [141] Y. Lu et al., “Software defined radio access network in 5G mobile
[119] H. Zhang et al., “5G wireless network: MyNET and SONAC,” IEEE network,” in Proc. 10th Int. Conf. Commun. Netw. China (ChinaCom),
Netw., vol. 29, no. 4, pp. 14–23, Jul./Aug. 2015. Shanghai, China, Aug. 2015, pp. 132–136.
[120] H. Huang, S. Guo, J. Wu, and J. Li, “Green datapath for TCAM-based [142] K. Chen et al., “HetSDN: Exploiting SDN for intelligent network usage
software-defined networks,” IEEE Commun. Mag., vol. 54, no. 11, in heterogeneous wireless networks,” in Proc. IEEE/ACM 24th Int.
pp. 194–201, Nov. 2016. Symp. Qual. Service (IWQoS), Beijing, China Jun. 2016, pp. 1–6.
[121] J. Wu et al., “Context-aware networking and communications: Part [143] R. Riggio, K. Gomez, L. Goratti, R. Fedrizzi, and T. Rasheed, “V-cell:
2 [guest editorial],” IEEE Commun. Mag., vol. 52, no. 8, pp. 64–65, Going beyond the cell abstraction in 5G mobile networks,” in Proc.
Aug. 2014. IEEE Netw. Oper. Manag. Symp. (NOMS), Kraków, Poland, May 2014,
[122] M. R. Sama, S. B. H. Said, K. Guillouard, and L. Suciu, “Enabling pp. 1–5.
network programmability in LTE/EPC architecture using OpenFlow,” [144] S. Sun, M. Kadoch, L. Gong, and B. Rong, “Integrating network func-
in Proc. 12th Int. Symp. Model. Optim. Mobile Ad Hoc Wireless tion virtualization with SDR and SDN for 4G/5G networks,” IEEE
Netw. (WiOpt), May 2014, pp. 389–396. Netw., vol. 29, no. 3, pp. 54–59, May/Jun. 2015.
[123] V.-G. Nguyen and Y. Kim, “Proposal and evaluation of SDN-based [145] M. Amani, T. Mahmoodi, M. Tatipamula, and H. Aghvami,
mobile packet core networks,” EURASIP J. Wireless Commun. Netw., “Programmable policies for data offloading in LTE network,” in Proc.
vol. 2015, no. 1, p. 172, 2015, doi: 10.1186/s13638-015-0395-1. IEEE Int. Conf. Commun. (ICC), Sydney, NSW, Australia, Jun. 2014,
[124] A. Ksentini, M. Bagaa, and T. Taleb, “On using SDN in 5G: The pp. 3154–3159.
controller placement problem,” in Proc. IEEE Glob. Commun. Conf. [146] “Huawei SoftCOM: Reshaping the future of network architec-
(GLOBECOM), Washington, DC, USA, Dec. 2016, pp. 1–6. ture,” Huawei, Shenzhen, China, Rep., 2013. [Online]. Available:
[125] J. Costa-Requena et al., “Software defined 5G mobile backhaul,” www.huawei.com/ilink/en/download/HW_204467
in Proc. 1st Int. Conf. 5G Ubiquitous Connectivity, Nov. 2014, [147] D. Sabella et al., “RAN as a service: Challenges of designing a flexible
pp. 258–263. RAN architecture in a cloud-based heterogeneous mobile network,”
[126] M. H. Kabir, “A novel architecture for SDN-based cellular network,” in Proc. Future Netw. Mobile Summit, Lisbon, Portugal, Jul. 2013,
Int. J. Wireless Mobile Netw., vol. 6, no. 6, pp. 71–85, Dec. 2014. pp. 1–8.
[127] X. An et al., “SDN-based vs. software-only EPC gateways: A cost [148] V. Jungnickel et al., “Software-defined open architecture for front-and
analysis,” in Proc. IEEE NetSoft Conf. Workshops (NetSoft), Seoul, backhaul in 5G mobile networks,” in Proc. 16th Int. Conf. Transp. Opt.
South Korea, Jun. 2016, pp. 146–150. Netw. (ICTON), Graz, Austria, Jul. 2014, pp. 1–4.
ZAIDI et al.: WILL SDN BE PART OF 5G? 3257

[149] P. Arnold, N. Bayer, J. Belschner, and G. Zimmermann, “5G radio [170] A. Hurtado-Borras, J. Palà-Solé, D. Camps-Mur, and S. Sallent-Ribes,
access network architecture based on flexible functional control / user “SDN wireless backhauling for small cells,” in Proc. IEEE Int. Conf.
plane splits,” in Proc. Eur. Conf. Netw. Commun. (EuCNC), Oulu, Commun. (ICC), London, U.K., Jun. 2015, pp. 3897–3902.
Finland, Jun. 2017, pp. 1–5. [171] R. Santos and A. Kassler, “A SDN controller architecture for small
[150] C. Yang, Z. Chen, B. Xia, and J. Wang, “When ICN meets C-RAN for cell wireless backhaul using a LTE control channel,” in Proc. IEEE
HetNets: An SDN approach,” IEEE Commun. Mag., vol. 53, no. 11, 17th Int. Symp. World Wireless Mobile Multimedia Netw. (WoWMoM),
pp. 118–125, Nov. 2015. Coimbra, Portugal, Jun. 2016, pp. 1–3.
[151] F. Han, S. Zhao, L. Zhang, and J. Wu, “Survey of strategies for [172] S. Sun, L. Gong, B. Rong, and K. Lu, “An intelligent SDN framework
switching off base stations in heterogeneous networks for greener 5G for 5G heterogeneous networks,” IEEE Commun. Mag., vol. 53, no. 11,
systems,” IEEE Access, vol. 4, pp. 4959–4973, 2016. pp. 142–147, Nov. 2015.
[152] X. Costa-Perez et al., “5G-Crosshaul: An SDN/NFV integrated [173] V. Sagar, R. Chandramouli, and K. P. Subbalakshmi, “Software defined
fronthaul/backhaul transport network architecture,” IEEE Wireless access for HetNets,” IEEE Commun. Mag., vol. 54, no. 1, pp. 84–89,
Commun., vol. 24, no. 1, pp. 38–45, Feb. 2017. Jan. 2016.
[153] H.-H. Cho, C.-F. Lai, T. K. Shih, and H.-C. Chao, “Integration of SDR [174] K.-K. Yap, “Blueprint for introducing innovation into wireless
and SDN for 5G,” IEEE Access, vol. 2, pp. 1196–1204, 2014. mobile networks,” in Proc. 2nd ACM SIGCOMM Workshop Virtual.
[154] E. J. Kitindi, S. Fu, Y. Jia, A. Kabir, and Y. Wang, “Wireless network Infrastruct. Syst. Archit. (VISA), 2010, pp. 25–32. [Online]. Available:
virtualization with SDN and C-RAN for 5G networks: Requirements, http://doi.acm.org/10.1145/1851399.1851404
opportunities, and challenges,” IEEE Access, vol. 5, pp. 19099–19115, [175] M. Yang et al., “Cross-layer software-defined 5G network,”
2017. Mobile Netw. Appl., vol. 20, no. 3, pp. 400–409, Jun. 2015,
[155] J. Mueller, Y. Chen, B. Reichel, V. Vlad, and T. Magedanz, “Design doi: 10.1007/s11036-014-0554-3.
and implementation of a carrier grade software defined telecommu- [176] K. Gomez, T. Rasheed, L. Reynaud, and L. Goratti, “FME: A flexible
nication switch and controller,” in Proc. IEEE Netw. Oper. Manag. management entity for virtualizing LTE evolved packet core,” in Proc.
Symp. (NOMS), May 2014, pp. 1–7. IEEE Netw. Oper. Manag. Symp. (NOMS), May 2014, pp. 1–4.
[156] J. Heinonen et al., “Dynamic tunnel switching for SDN-based [177] M. Yang et al., “Software-defined and virtualized future mobile and
cellular core networks,” in Proc. ACM 4th Workshop All Things wireless networks: A survey,” Mobile Netw. Appl., vol. 20, no. 1,
Cellular Oper. Appl. Challenges (AllThingsCellular), Chicago, IL, pp. 4–18, Feb. 2015, doi: 10.1007/s11036-014-0533-8.
USA, 2014, pp. 27–32. [Online]. Available: http://doi.acm.org/ [178] N. A. Jagadeesan and B. Krishnamachari, “Software-defined
10.1145/2627585.2627587 networking paradigms in wireless networks: A survey,” ACM Comput.
[157] J. Pagé and J.-M. Dricot, “Software-defined networking for low- Surveys, vol. 47, no. 2, pp. 1–11, Jan. 2015. [Online]. Available:
latency 5G core network,” in Proc. Int. Conf. Military Commun. Inf. http://doi.acm.org/10.1145/2655690
Syst. (ICMCIS), May 2016, pp. 1–7. [179] S. Tomovic, M. Pejanovic-Djurisic, and I. Radusinovic, “SDN based
[158] S. Chourasia and K. M. Sivalingam, “SDN based evolved packet core mobile networks: Concepts and benefits,” Wireless Pers. Commun.,
architecture for efficient user mobility support,” in Proc. 1st IEEE Conf. vol. 78, no. 3, pp. 1629–1644, 2014, doi: 10.1007/s11277-014-1909-6.
Netw. Softwarization (NetSoft), London, U.K., Apr. 2015, pp. 1–5. [180] J. Li, P. Liu, and H. Li, “Software-defined cellular mobile network
[159] “Relationship of SDN and NFV, issue 1,” Open Netw. Found., solutions,” ZTE Commun., vol. 12, no. 2, pp. 28–33, Jun. 2014.
Menlo Park, CA, USA, Rep. TR-518, Oct. 2015. [Online]. Available: [181] T. Chen, M. Matinmikko, X. Chen, X. Zhou, and P. Ahokangas,
https://www.opennetworking.org/images/stories/downloads/sdn- “Software defined mobile networks: Concept, survey, and research
resources/technical-reports/onf2015.310_Architectural_comparison.08- directions,” IEEE Commun. Mag., vol. 53, no. 11, pp. 126–133,
2.pdf Nov. 2015.
[160] “Network function virtualisation (NFV); ecosystem; report on SDN [182] I. T. Haque and N. Abu-Ghazaleh, “Wireless software defined
usage in NFV architectural framework,” Eur. Telecommun. Stand. Inst., networking: A survey and taxonomy,” IEEE Commun. Surveys Tuts.,
Sophia Antipolis, France, Rep. GS NFV-EVE 005 V1.1.1, Dec. 2015. vol. 18, no. 4, pp. 2713–2737, 4th Quart., 2016.
[Online]. Available: http://www.etsi.org/deliver/etsi_gs/NFV- [183] V.-G. Nguyen, A. Brunstrom, K.-J. Grinnemo, and J. Taheri,
EVE/001_099/005/01.01.01_60/gs_NFV-EVE005v010101p.pdf “SDN/NFV-based mobile packet core network architectures: A sur-
[161] F. Z. Yousaf, M. Bredel, S. Schaller, and F. Schneider, “NFV and vey,” IEEE Commun. Surveys Tuts., vol. 19, no. 3, pp. 1567–1602,
SDN—Key technology enablers for 5G networks,” IEEE J. Sel. Areas 3rd Quart., 2017.
Commun., vol. 35, no. 11, pp. 2468–2478, Nov. 2017. [184] M. A. Al-Namari, A. M. Mansoor, and M. Y. I. Idris, “A brief survey
[162] 5G NORMA, “Deliverable D3.2, 5G norma network architecture— on 5G wireless mobile network,” Int. J. Adv. Comput. Sci. Appl., vol. 8,
Intermediate report,” Eur. Res. Council, Huntsville, AL, USA, Rep. no. 11, pp. 52–59, 2017, doi: 10.14569/IJACSA.2017.081107.
H2020-ICT-2014-2 5G NORMA/D3.2, Jan. 2017. [Online]. Available: [185] S. K. Tayyaba and M. A. Shah, “5G cellular network integration with
https://5gnorma.5g-ppp.eu/wp-content/uploads/2017/03/5g_norma_d3- SDN: Challenges, issues and beyond,” in Proc. Int. Conf. Commun.
2.pdf Comput. Digit. Syst. (C-CODE), Mar. 2017, pp. 48–53.
[186] D. A. Temesgene, J. Núñez-Martínez, and P. Dini, “Softwarization and
[163] M. R. Sama et al., “Software-defined control of the virtualized mobile
optimization for sustainable future mobile networks: A survey,” IEEE
packet core,” IEEE Commun. Mag., vol. 53, no. 2, pp. 107–115,
Access, vol. 5, pp. 25421–25436, 2017.
Feb. 2015.
[164] F. Z. Yousaf, J. Lessmann, P. Loureiro, and S. Schmid, “SoftEPC— Zainab Zaidi (M’01–SM’14) received the Bachelor
Dynamic instantiation of mobile core network entities for efficient of Engineering degree from the NED University of
resource utilization,” in Proc. IEEE Int. Conf. Commun. (ICC), Engineering and Technology, Karachi, Pakistan, in
Budapest, Hungary, Jun. 2013, pp. 3602–3606. 1997 and the Ph.D. degree from George Mason
[165] E. Haleplidis et al., “Building softwarized mobile infrastructures with University, VA, USA, in 2004, focused on track-
ForCES,” in Proc. 23rd Int. Conf. Telecommun. (ICT), Thessaloniki, ing mobile callers in a wireless system. She was a
Greece, May 2016, pp. 1–5. Visiting Senior Research Fellow with King’s College
[166] V. Ziegler et al., “Architecture vision for the 5G era,” in Proc. IEEE Int. London, U.K., on Australian Endeavour Fellowship
Conf. Commun. Workshops (ICC), Kuala Lumpur, Malaysia, May 2016, in 2014. She was with an Australian ICT research
pp. 51–56. organization NICTA from 2006 to 2013. Her work
[167] L. Valtulina, M. Karimzadeh, G. Karagiannis, G. Heijenk, and A. Pras, was published in top ACM and IEEE conferences.
“Performance evaluation of a SDN/OpenFlow-based distributed mobil- The major results were published in the IEEE T RANSACTIONS ON M OBILE
ity management (DMM) approach in virtualized LTE systems,” in C OMPUTING. She has extensive experience in using statistical methods,
Proc. IEEE Globecom Workshops (GC Wkshps), Austin, TX, USA, machine learning tools, network simulators, and wireless testbeds. She has
Dec. 2014, pp. 18–23. over 30 publications, mostly as first and second author, and has over 584
[168] S. Kuklinski, Y. Li, and K. T. Dinh, “Handover management in SDN- citations according to Google Scholar. She have supervised research works of
based mobile networks,” in Proc. IEEE Globecom Workshops (GC various M.S. students and also co-supervised couple of Ph.D. students. Her
Wkshps), Austin, TX, USA, Dec. 2014, pp. 194–200. research interests mainly toward new networking paradigms, such as, SDN,
[169] Z. Cao, S. S. Panwar, M. Kodialam, and T. V. Lakshman, “Enhancing separation of control and data for energy efficiency, etc. She is interested in
mobile networks with software defined networking and cloud com- mathematical characterization of networking architectures in order to quan-
puting,” IEEE/ACM Trans. Netw., vol. 25, no. 3, pp. 1431–1444, tify performance limits. She served as a TPC member of conferences and
Jun. 2017. workshops, such as IEEE ICCCN and ACM MOBIWAC.
3258 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 4, FOURTH QUARTER 2018

Vasilis Friderikos is currently a Reader with the Mischa Dohler is a Full Professor in wireless
Centre for Telecommunications Research, King’s communications with King’s College London, driv-
College London. He has been a Visiting Researcher ing cross-disciplinary research and innovation in
with WinLab, Rutgers University, USA. He has technology, sciences, and arts. He is a fellow of
supervised 9 Ph.D. students published over 200 the Royal Academy of Engineering, the Royal
research papers. His research interests lie broadly Society of Arts, the Institution of Engineering and
within the closely overlapped areas of wireless Technology, and a Distinguished Member of Harvard
networking, mobile computing, and architectural Square Leaders Excellence. He acts as a Policy
aspects of the Future Internet. He has also been Advisor on issues related to digital, skills and edu-
teaching advanced mobility management protocols cation. He has had ample coverage by national and
for the Future Internet with the Institut Supérieur international press and media.
de l’Electronique et du Numérique, France, in 2010. He was a recipient of He is a frequent keynote, panel and tutorial speaker, and has received
best paper awards in IEEE ICC 2010 and WWRF conferences two times and numerous awards. He has pioneered several research fields, contributed to
the British Telecom Fellowship Award in 2005. He has been a Organizing numerous wireless broadband, IoT/M2M and cyber security standards, holds
Committee Member of the Green Wireless Communications and Networks a dozen patents, organized and chaired numerous conferences. He was the
Workshop during VTC Spring 2011. He has been the Track Co-Chair at Editor-in-Chief of two journals, has over 200 highly-cited publications, and
the IEEE VTC-2015 Spring, IEEE PIMRC 2013, WWRF-32, and the IEEE authored several books.
WCNC 2010 Conference (acting as a Technical Program Committee Member Mr. Dohler was the Director of the Centre for Telecommunications
for IEEE Globecom, IEEE ICC and 40 other flagship international conferences Research, King’s College London from 2014 to 2018. He is the Co-Founder
over the last seven years). He is a member of IET and INFORMS Section on of the Smart Cities pioneering company Worldsensing, where he was the
Telecommunications. CTO from 2008 to 2014. He also worked as a Senior Researcher with
Orange/France Telecom from 2005 to 2008.

Zarrar Yousaf received the B.Sc. and M.Sc.

degrees in EE from the University of Engineering
and Technology, Peshawar, Pakistan, in 1996 and
1999, respectively, the M.Sc. degree in telecommu-
nication and computers from George Washington
University, USA, in 200, and the Ph.D. degree
from the Dortmund University of Technology
(TU Dortmund), Germany, in 2010. He is a
Senior Researcher with NEC Laboratories Europe,
Heidelberg, Germany. His current research interest
is focused around NFV/SDN related technologies in
the context of 5G network architecture and operations. He is also a Delegate at
the ETSI NFV standards organization, where he has contributed to around sev-
eral published ETSI NFV standard documents and holding Rapporteurship for
multiple work items. He has extensive experience in the design development, Hamid Aghvami joined King’s College London as
modeling, simulation, and prototyping of novel protocols, algorithms, and an Academic Staff in 1984, where he was promoted
system methods for optimizing network performance. He has 12 filed patents. to a Reader in 1989, and in 1993 was promoted to a
His research work has been published in several peer reviewed international Professor in telecommunication engineering. He is
journals, conferences, book chapters, and Internet drafts. the Founder of the Centre for Telecommunications
Research, King’s College London, where he was
the Director from 1994 to 2014.
He carries out consulting work on Digital
Radio Communications Systems for British and
Simon Fletcher joined Real Wireless in 2016 as International companies. He has published over
a Chief Technology Officer, taking responsibility 550 technical journal and conference papers, filed
for overall technical and innovation strategy. He 30 patents and given invited talks and courses the world over on various
has chaired various conference committees, includ- aspects of Mobile Radio Communications. He was a Visiting Professor
ing the Cambridge Wireless Annual International with NTT Radio Communication Systems Laboratories in 1990, a Senior
Conference and is recognized as a thought leader Research Fellow with BT Laboratories from 1998 to 1999, and was an
regularly speaking at industry events. His long- Executive Advisor to Wireless Facilities Inc., USA, from 1996 to 2002. He
standing association with the U.K. innovation eco- is the Chairman of Advanced Wireless Technology Group Ltd. He is also the
system through roles such as the Director of mVCE Managing Director of Wireless Multimedia Communications Ltd., his own
and the Innovate-U.K. ICT-KTN brings a wealth consultancy company.
of practical knowledge on the application of open Prof. Aghvami leads an active research team working on numerous
innovation to accelerate product, service and process delivery. He has spent mobile and personal communications projects for fourth and fifth generation
20 years working in telecoms infrastructure systems and product develop- networks; these projects are supported both by government and industry.
ment. In NEC Corporation, he played a key role in the formation of Joint He was a member of the Board of Governors of the IEEE Communications
Ventures for the development of 3G and 4G products and established a Core Society from 2001 to 2003. He was a Distinguished Lecturer of the IEEE
Team that developed the first-generation of technology for 4G systems, ulti- Communications Society from 2004 to 2007. He has been a member, the
mately representing NEC in a Steering Board position in the LTE SAE Trials Chairman, and the Vice-Chairman of the technical programme and organizing
Initiative during the early commercial product launch. His long participation committees of a large number of international conferences. He is also the
in Common Public Radio Interface defining early C-RAN concepts brings Founder of the International Symposium on Personal Indoor and Mobile
foresight on an important architectural element of emerging 5G architectures. Radio Communications, a major yearly conference attracting some 1000
He pioneered strategic engagements on behalf of NEC in the area of future attendees.
cities, the application of 5G and IoT in industry verticals with an event horizon His research achievements as the Director of the Centre for
toward 2030. He currently directs the Real Wireless engagement in H2020 Telecommunications Research, King’s College London. He was a
5G-NORMA (also serving at the project level as the Innovation Manager), 5G- recipient of the IEEE Technical Committee on Personal Communications
MoNArch and UK5G AutoAir projects overseeing the establishment of new Recognition Award in 2005 for his outstanding technical contributions to the
techno-economic analysis capabilities for 5G. He led a study for the U.K.’s communications field, and for his service to the scientific and engineering
National Infrastructure Commission on Future Communications infrastructure communities and the Fellowship of the Wireless World Research Forum in
use cases and costs. He is a member of the recently formed UK5G Advisory recognition of his personal contributions to the wireless world in 2009. He
Board. is a fellow of the Royal Academy of Engineering and IET.