SPECIAL SECTION ON MOBILE EDGE COMPUTING AND MOBILE CLOUD COMPUTING:

ADDRESSING HETEROGENEITY AND ENERGY ISSUES OF COMPUTE AND NETWORK RESOURCES

Received August 2, 2019, accepted August 26, 2019, date of publication August 30, 2019, date of current version September 18, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2938534

Edge Computing in 5G: A Review

NAJMUL HASSAN 1 , KOK-LIM ALVIN YAU 1 , (Senior Member, IEEE),
AND CELIMUGE WU 2 , (Senior Member, IEEE)
1 Department of Computing and Information Systems, Sunway University, Bandar Sunway 47500, Malaysia
2 Graduate School of Informatics and Engineering, The University of Electro-Communications, Tokyo 182-8585, Japan
Corresponding author: Kok-Lim Alvin Yau ([email protected])

ABSTRACT 5G is the next generation cellular network that aspires to achieve substantial improvement on
quality of service, such as higher throughput and lower latency. Edge computing is an emerging technology
that enables the evolution to 5G by bringing cloud capabilities near to the end users (or user equipment,
UEs) in order to overcome the intrinsic problems of the traditional cloud, such as high latency and the lack of
security. In this paper, we establish a taxonomy of edge computing in 5G, which gives an overview of existing
state-of-the-art solutions of edge computing in 5G on the basis of objectives, computational platforms,
attributes, 5G functions, performance measures, and roles. We also present other important aspects, including
the key requirements for its successful deployment in 5G and the applications of edge computing in 5G. Then,
we explore, highlight, and categorize recent advancements in edge computing for 5G. By doing so, we reveal
the salient features of different edge computing paradigms for 5G. Finally, open research issues are outlined.

INDEX TERMS 5G, cloud computing, edge computing, fog computing.

I. INTRODUCTION and c) make and distribute decisions on mini clouds locally.

Edge computing is a computational paradigm that enables Hence, edge servers in the mini clouds have the capabilities
edge servers in mini clouds (or edge clouds) to extend of a cloud but on a different scale, and they are located locally
cloud capabilities at the edge of the network to perform instead of remote data centers which may be far away from
computationally-intensive tasks and store a massive amount UEs [9].
of data at close proximity to user equipment (UEs) [1]–[3].
Traditional cloud computing, which is a centralized com- A. OUR CONTRIBUTIONS
puting paradigm that provides continuous access to highly This paper highlights recent advances of edge comput-
capable data centers, has been adopted to allow UEs to ing in 5G. Some analyses have been made on a par-
offload computation and storage to the data centers [4]. This ticular computational platform of edge computing in 5G,
is because UEs have limited processing, computational, and particularly mobile edge computing (MEC) [10], [11].
storage capabilities. Nevertheless, edge computing is pre- In addition [12], [13] focuses on edge orchestration and its
ferred to cater for the wireless communication requirements related issues in 5G environment and MEC architecture.
of next generation applications, such as augmented reality In [14] a survey of edge computing, including its applications
and virtual reality, which are interactive in nature. These and key challenges, is presented from the perspective of
highly interactive applications are computationally-intensive vehicular networks. Our paper is first of its kind to present: a)
and have high quality of service (QoS) requirements, includ- a taxonomy of edge computing in 5G covering the objectives,
ing low latency and high throughput (e.g, ultra reliable low computational platforms, attributes, the use of 5G functions,
latency communication (URLLC), tactile internet) [5]–[7]. performance measures, and the role of edge computing; b)
Most importantly, these applications are expected to generate a review of state-of-the-art edge computing schemes in 5G;
a massive amount of data up to 30.6 exabytes per month [8]. and c) open issues in this research topic. This topic is timely
The limited capabilities of UEs warrants the need for edge due to the recent advent of 5G, and the evolving roles of edge
computing to: a) receive and store a massive amount of computing in the realization of 5G.
real-time data, b) process, compute, and analyze the data,
B. ORGANIZATION OF THIS PAPER
The associate editor coordinating the review of this manuscript and The rest of this paper is organized as follows: Section II
approving it for publication was Junaid Shuja. presents an overview of 5G and edge computing, respectively,

127276 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ VOLUME 7, 2019
N. Hassan et al.: Edge Computing in 5G: Review

and answers a host of questions on the use of edge computing Applications, such as intelligent traffic signal control
in 5G, including the time characteristics of data, the key system, generate this kind of data.
requirements, and the applications of edge computing in 5G. • Non-real-time data is not time-sensitive and can tolerate
Section III presents a taxonomy of edge computing in 5G. latency.
Section IV presents the state-of-the-art schemes for edge Edge computing is envisioned to handle applications and ser-
computing in 5G. Section V presents open research issues. vices with hard real-time requirement using edge servers due
Section VI concludes this paper. to their close proximity to UEs leading to significant reduc-
tion in latency. For applications and services with soft-real
II. BACKGROUND time requirement, or bounded end-to-end delay, tasks are han-
This section presents an overview of 5G, the time characteris- dled by edge servers if the response delay between UEs and
tics of data, edge computing, as well as the key requirements the cloud is higher than the requirement; otherwise, the tasks
and the applications of edge computing in 5G. can be offloaded to the cloud. For applications and services
with non-real-time requirement, tasks can be offloaded to the
A. REQUIREMENTS OF 5G SYSTEMS cloud for load balancing.
5G is foreseen as the next generation wireless cellular net-
work to cater for the needs of next generation networks. C. SIGNIFICANCE OF EDGE COMPUTING
5G possesses three main characteristics unseen in previous 5G is foreseen to support highly interactive applications with
generation networks. Firstly, a massive amount of data is gen- low latency and high throughput requirements [21]. Edge
erated. According to the International Telecommunication computing adopts a decentralized model that brings cloud
Union (ITU), there are more than 7.5 billion mobile devices computing capabilities closer to UEs in order to reduce
around the world in 2017 [15], and the number of mobile latency. Fig. 1 shows the cloud computing and edge comput-
devices is expected to increase to 25 billion by 2020 [16], ing models. Edge computing can either operate as a single
contributing to ultra-dense networks. Consequently, there is computing platform, or a collaborative platform together with
an explosive growth in the amount of data from 16.5 exabytes other components, including the cloud [22]. Edge comput-
in 2014 to an estimate of 500 exabytes in 2020 [17], con- ing is necessary as the traditional cloud computing model
tributing to a growth rate of 30 times. Secondly, stringent is not suitable for highly interactive applications that are
QoS requirements are imposed to support highly interactive computationally-intensive and have high QoS requirements,
applications, requiring ultra-low latency and high throughput. including low latency and high throughput. This is because
Thirdly, heterogeneous environment must be supported to cloud may be far away from UEs, which also increases
allow inter-operability of a diverse range of UEs (e.g., smart
phones and tablets), QoS requirements (e.g., different levels
of latency and throughput for multimedia applications), net-
work types (e.g., IEEE 802.11 and Internet of things), and
so on.
5G is comprised of three main new technologies to provide
higher network capacity in order to support a higher number
of UEs [18]. Firstly, mmWave communication, which uses
high frequency bands (i.e., 30 GHz to 300 GHz [19]), pro-
vides high bandwidth (i.e., at least 11 Gbps [20]). Secondly,
small cells deployment allows UEs to communicate using
mmWave in order to reduce transmission range and inter-
ference. Thirdly, massive MIMO (multiple-input multiple-
output) allows base stations (BSs) to use a large number
of antennas (e.g., up to 16 antennas per sector) to provide
directional transmission (or beamforming) in order to reduce
interference, allowing neighboring nodes to communicate
simultaneously.

B. MAIN CHARACTERISTICS OF 5G DATA

Data can be categorized into three main categories according
to its time characteristics as follows:
• Hard real-time data has a strict predefined latency.
Applications, such as video steaming, gaming, and
healthcare services, generate this kind of data.
• Soft real-time data has a predefined latency, yet it
can tolerate some pre-defined and bounded latency. FIGURE 1. Cloud computing and edge computing models.

VOLUME 7, 2019 127277

 N. Hassan et al.: Edge Computing in 5G: Review

energy consumption. In other words, cloud servers are typ- • Entertainment and multimedia applications, such as
ically located at the core network, and edge servers of the streaming HDTV or 3D TV.
mini clouds are located at the edge of the network [23]. • Virtual reality, augmented reality, and mixed reality,
To understand the need of an edge computing, con- such as streaming video contents to virtual reality
sider real-time packet delivery among self-driving cars that glasses. The size of the glasses can be reduced by
requires an end-to-end delay of less than 10 ms [24]. The offloading computation from the glasses to edge servers.
minimum end-to-end delay for an access to cloud is greater • Tactile internet, which is the next evolution of Internet
than 80 ms [25], [26], which is intolerable. Edge computing of things, provides an ultra-responsive and ultra-reliable
fulfills the sub-millisecond requirement of 5G applications, network connectivity to ensure successful delivery of
and reduces energy consumption by around 30% to 40% [27], real-time control messages and physical tactile experi-
which attributes to up to five times lesser energy consumption ences remotely [32], [33].
as compared to accessing the cloud [28]. • URLLC, which ensures high reliability between
UEs specifically in M2M communications, supports
D. KEY REQUIREMENTS OF EDGE COMPUTING IN 5G low-latency transmissions of small payloads with very
There are four key requirements for the successful deploy- high reliability from a limited set of UEs, such as fire
ment and operation of edge computing in 5G. While all the alarms [34].
four key requirements are important, achieving a balanced • Internet of things, such as smart appliances that connect
trade-off among them must be considered depending on the devices (e.g., household appliances) to the internet.
applications. • Factories of the future, such as smart machines,
Firstly, real-time interaction, which is the fundamen- to improve safety and productivity. Operators can use a
tal motivation for the use of edge computing over cloud remote platform to operate heavy machines, particularly
computing, ensures low latency to support delay-sensitive those located at hard-to-reach and unsafe places, from a
applications and services (e.g. remote surgery, tactile inter- safe and comfortable place.
net, URLLC, unmanned vehicles [29], [30] and vehicle • Emergency response, whereby different kinds of data
accident prevention) in order to improve QoS. A diverse and information about an event or incident are gathered
range of services, including decision making and data from different sources at different times. The partially
analysis, can be provided by edge servers in a real-time available data and information are used to make critical
manner. decisions, and they provide a more complete picture of
Secondly, local processing is feasible since data and user the event as time goes by. Decisions made are shared
requests can be processed by edge servers, rather than the with emergency response team (e.g., firefighters) in real
cloud. This means that, by reducing the traffic amount across time, even prior to their arrivals at the location of the
the connection between a small cell and the core network: event.
a) the bandwidth of the connection can be increased to prevent • Intelligent transportation system, whereby drivers can
bottleneck; and b) the traffic amount in the core network is share or gather information from traffic information
reduced. centers to avoid vehicles that are in danger, or stop
Thirdly, high data rate is necessary to transmit the massive abruptly, in a real-time manner in order to avoid
amount of data generated by a diverse range of applications accidents. In addition, unmanned vehicles can sense
(e.g., virtual reality and remote surgery) to edge clouds [31]. their surroundings and move safely in an autonomous
Edge servers, which can be embedded in the BSs, allow easy manner.
access to edge clouds without the need to access the core
network. The use of mmWave frequency bands in a small cell III. TAXONOMY
provides a high data rate transmission. Fig. 2 shows a taxonomy of edge computing in 5G, covering
Fourthly, high availability ensures the availability of the objectives, computational platforms, attributes, the use of 5G
cloud services at the edge. Since edge computing pushes data functions, performance measures, and the roles of edge com-
and application logic to the edge clouds, the availability of puting in 5G.
the edge clouds is important.
A. OBJECTIVES
E. APPLICATIONS OF EDGE COMPUTING IN 5G There are five main objectives of edge computing in 5G as
Many applications of 5G are relying on edge computing for follows:
real-time interaction, local processing, high data rate, and O.1 Improving data management to handle a large amount of
high availability, including: delay-sensitive data, which are generated by UEs, that
• Healthcare, such as remote surgery and diagnostics, needs to be handled locally in a real-time manner. For
as well as monitoring of patient vital signs and data. instance, the local UEs in a smart factory is expected
Doctors can use a remote platform to operate surgical to generate up to 1 petabyte of data daily [35]. Since
tools in order to save life from a distance where they accessing to cloud incurs high latency [36], the data
feel safe and comfortable. can be handled locally by edge servers. Such efficient

127278 VOLUME 7, 2019

N. Hassan et al.: Edge Computing in 5G: Review

FIGURE 2. Taxonomy of edge computing in 5G.

data management is needed to support local functions to outsource services and data to edge clouds. For
(e.g., D2D) and real-time applications (e.g., remote instance, mobile UEs can query and search for informa-
surgery). tion about points of interest in local proximities given
O.2 Improving QoS to meet a diverse range of stringent QoS their geographical locations. The number of queries can
requirements in order to improve quality of experience be high, such as queries related to hospitals and medical
(QoE) [37]. This helps to support next generation appli- advices during emergency response.
cations, including highly interactive applications and O.5 Improving resource management to optimize network
on-demand services. For instance, over-the-top (OTT) resource utilization for network performance enhance-
services enable online delivery of multimedia contents, ment due to the limited network resources available
which generally require low latency and high bandwidth, in the edge cloud as compared to the cloud. This is
without the service providers being actively involved in challenging as it is a multi-objective function that must
the control and distribution of the content [38]. This cater to a diverse range of applications, as well as user
service can promote new and personalized applications requirements and demands, which vary as time goes by.
that allow service providers to customize QoS [39]. Ser-
vice providers must have a holistic view of subscribers B. COMPUTATIONAL PLATFORMS
and customers, covering their contextual information, Different computational platforms provide varying comput-
such as their preferences and interests. Subsequently, ing capabilities (e.g., in terms of processing loads) with differ-
the information can be personalized for attracting poten- ent characteristics (e.g., in terms of availability, the proximity
tial customers and enhancing their QoE. from UEs, and the complexity of network infrastructure) to
O.3 Predicting network demand to estimate the required net- process data at different geographical locations. The com-
work resources to cater for the network (or user) demand putational platform can be used either individually or in
in a local proximity, and subsequently to provide optimal combination based on the network scenarios and application/
resource allocation to handle the local network demand. service requirements. As an example for the computational
An accurate prediction of network demand helps to platform used individually, applications and services with
decide whether a network demand should be handled strict QoS requirements can use edge servers to process
locally at the edge or at the cloud, and so it provides an real-time data. As an example for the computational platform
efficient allocation of resources (e.g., bandwidth). used in combination, healthcare applications and services
O.4 Managing location awareness to enable the geographi- with both real-time and non-real-time data can use edge
cally distributed edge servers to infer their own locations servers to process real-time and lightweight data, and cloud
and track the location of UEs to support location-based to process heavyweight data. There are three main computa-
services. This enables location-based service providers tional platforms in 5G as follows:

VOLUME 7, 2019 127279

 N. Hassan et al.: Edge Computing in 5G: Review

TABLE 1. Comparison between fog and MEC.

C.1 Cloud computing gathers, processes, and stores a mas- processes real-time data and makes real-time decisions,
sive amount of network-wide data and information from while cloud computing processes non-real-time data
UEs in the network. Subsequently, it sends back the data and makes non-real-time decisions. The hybrid infras-
and information, or decisions, back to the UEs. While tructure combines the advantages of both edge com-
cloud can empower UEs with low computational and puting (i.e., real-time responses) and cloud computing
storage capabilities, it is not suitable to provide real-time (i.e., high computational and storage capabilities). Com-
services because cloud may be far away from UEs. putation can be performed in different layers, particu-
C.2 Edge computing gathers, processes, and stores a massive larly the cloud (or the upper) layer and the fog or the
amount of local data and information from UEs in a local edge (or the bottom) layer. In general, real-time tasks are
area. Edge computing has close proximity to UEs, while executed in the fog layer, and tasks requiring high com-
cloud may be far away from UEs. Table 1 presents a putation are executed in the cloud layer. Compared to the
comparison of three types of edge computing platforms traditional cloud, edge computing increases throughput
as follows: and reduces latency, which are important to support
C.2.1 Fog computing deploys local fog nodes which delay-sensitive applications. Nevertheless, the hybrid
are local hardware devices, such as switches and platform is more complex compared to the separate
routers, to provide local computation. Accord- cloud computing and edge computing platforms.
ing to the OpenFog Consortium, fog computing
is ‘‘a system-level horizontal architecture that C. ATTRIBUTES
distributes resources and services of computing, Edge computing has three main attributes as follows:
storage, control, and networking anywhere along T.1 Low latency and close proximity enables edge computing
the continuum from cloud to things’’ [40]. Fog to reduce the response delay (or round-trip time) suffered
computing shares similar benefits to other edge by UEs while accessing the traditional cloud. There are
computing variants (e.g., MEC) to provide low three main components in a response delay: a) communi-
latency and real-time analytics; however, it has cation delay that depends on data rate; b) computational
low storage capacity. delay that depends on computational time; and c) prop-
C.2.2 MEC provides storage and computational capac- agation delay that depends on propagation distance.
ities at the edge of the network, such as the radio In general, in cloud computing, the end-to-end delay is
access networks (RANs) and BSs, to improve greater than 80ms (or 160ms for response delay) [25].
context awareness and reduce latency. The MEC This is not suitable for delay-sensitive applications, such
servers, which are usually co-located with mul- as remote surgery and VR, that require tactile speed with
tiple hosts (e.g., BSs), use a virtualized inter- a response delay of at most 1ms [41]. In edge comput-
face to access storage and computation facilities. ing, UEs experience reduced overall end-to-end delay
A MEC orchestrator overlooks the MEC hosts and response delay due to their close proximity to edge
by gathering and providing real-time informa- servers. The strategic location of edge cloud reduces the
tion regarding the services offered by each host, communication and propagation delays. For instance,
the available resources (e.g., network capacity the propagation distance is reduced to tens of meters
and load), the network topology (e.g., UEs con- via D2D communication and in small cells, and it is
nected to the servers including their location and generally limited within a kilometer from the UEs [42].
networking information), as well as managing T.2 Location awareness enables edge servers to collect and
MEC applications. process data generated by UEs on the basis of the geo-
C.1 Hybrid combines cloud computing and edge computing graphical location of UEs. This allows location-based
so that they can cooperate. For instance, edge computing and personalized service provisioning to UEs, whereby

127280 VOLUME 7, 2019

N. Hassan et al.: Edge Computing in 5G: Review

edge servers can gather data generated by sources in its without passing through BSs, which improves sys-
proximity without sending it to the cloud. tem throughput, energy efficiency, and spectrum
T.3 Network context awareness enables edge servers to utilization [50], [51]. UEs can offload tasks and com-
acquire network context information. This is because putations to edge servers in order to empower UEs with
edge servers tend to possess network context infor- computational capabilities. This feature of edge comput-
mation, particularly the real-time network conditions ing ensure successful D2D communication [52], [53].
(e.g., traffic load in a network cell, and radio access
network information) and UEs’ information (e.g., allo- E. ROLES OF EDGE COMPUTING IN 5G
cated bandwidth and user location). The information There are six main roles of edge computing to sup-
allows edge servers to adapt and respond to the vary- port real-time and interactive applications and services as
ing network conditions and UEs, and subsequently to follows:
optimize network resource utilization. This helps edge R.1 Local storage. Edge computing offloads a massive
servers to handle a massive amount of traffic in order to amount of data from UEs to edge clouds. While edge
improve network performance. Fine-granular informa- servers offer distributed local storage for a significant
tion (e.g., precise individual resource reservation infor- amount of data, yet their storage is much lower than
mation) can also be used to provide specific services that in the cloud, which has virtually unlimited storage
to traffic flows in order to cater for individual user capacity. Examples of data being stored are computing
requirements. strategies (e.g., computation offloading strategy [54]),
metadata (e.g., timestamps and geographical locations),
D. USE OF 5G FUNCTIONS and monitoring data. The edge server provides dif-
Five major enablers of edge computing in 5G are as ferent types of storage strategies to support different
follows: kinds of data. For instance, ephemeral storage provides
U.1 Software-defined network (SDN) defines a network temporary data storage to a set of interconnected mobile
architecture that separates a network into control and devices [55], [56].
data planes to provide flexible and agile networks, which R.2 Local computation. Edge computing offloads computa-
helps to simplify network management and deploy new tion and process from less complex (e.g., smart phone)
services [43]. In general, the control plane handles and highly complex (e.g., surgical tools and smart fac-
policy on cloud, while the data plane forwards traf- tories) UEs to edge clouds. While traditional cache and
fic according to decisions made by the control plane. access technologies (e.g., IEEE 802.11) provide simple
Network functions (e.g., routing) that require real-time computation, the edge cloud is an intelligent computing
response can be handled by edge servers [44]. system that provides local computation and data pro-
U.2 Network function virtualization (NFV) performs net- cessing capabilities close to UEs in an independent and
work functions (also known as virtual functions) in vir- autonomous manner [57]. The outcomes of the compu-
tual machines on servers, which can handle a massive tations and processes can be valuable inputs to other
amount of data to provide flexible, automated, and scal- UEs, such as those in a smart factory. The advantage
able networks [45], [46]. Network demands can be pro- is that edge clouds perform small tasks and provide
cessed either at the cloud or at the edge, which prevent real-time responses locally, which help to reduce the
all data and information from being sent to the cloud. cost and delay incurred to send the required data to the
U.3 Massive MIMO deploys multiple antenna elements cloud.
to increase an antenna array at transmitter and R.3 Local data analysis. Edge computing processes and per-
receiver. This is in accordance to the Shannon theo- forms critical and real-time data analysis on a massive
rem [47] in which the signal-to-noise ratio increases amount of raw data gathered from different applications
without the need to increase the transmission power, in close proximity to generate valuable information [58].
leading to increased network capacity and energy effi- The capability to make data analysis locally reduces
ciency. Using massive MIMO, multiple UEs can offload the latency required to send data to, as well as to wait
tasks to an edge server simultaneously to reduce latency for responses from, the cloud. Subsequently, the out-
and energy consumption [48]. comes of the local data analysis are used for decision
U.4 Dynamic access to radio access technologies provide making [59].
access to conventional access technologies, such as R.4 Local decision making. Edge computing helps enti-
Wi-Fi, and new radio access technology (RAT) in 5G, ties to make real-time decisions and corresponding
such as NR [49]. 5G NR is a new standard that provides actions in an automated manner based on well-processed
connection to a diverse range of devices for achieving data [60]. The capability to make decisions locally
low latency and scalable networks, which allow future reduces involvement from more components and data
extension to existing networks. or information exchange, leading to: a) improved sys-
U.5 D2D communication enables direct communica- tem availability, particularly the cloud; and b) improved
tion between neighboring UEs using ad-hoc links bandwidth availability. As an example, edge computing

VOLUME 7, 2019 127281

 N. Hassan et al.: Edge Computing in 5G: Review

TABLE 2. summary of objectives, challenges, metrics, characteristics, and performance measures of clustering schemes for 5G networks.

facilitates local decision making by automated factories. P.2 Higher QoS. Edge computing improves QoS by pro-
Multiple entities can make decisions in a collaborative viding local functions (R.1)-(R.5). This reduces the
manner. amount of tasks and data offloaded to the cloud,
R.5 Local operation. Edge computing enables remote con- and so it increases network performance (e.g., higher
trol and monitoring – particularly critical devices includ- throughput and lower latency), which are important to
ing those under unsafe environment – from a distance, or delay-sensitive applications (e.g, remote surgery and
a more comfortable or safer place [61]. online gaming).
R.6 Local security enhancement. Edge computing serves as P.3 Energy efficiency reduces energy consumption by pro-
an additional layer between the cloud and connected viding local functions (R.1)-(R.5). This reduces the
devices in order to improve network security, includ- amount of energy incurred to offload tasks and data to
ing UEs with limited resources [62]. The edge clouds the cloud (i.e., the energy incurred in communication),
can serve as secured distributed platforms that provide and so it increases network lifetime.
security credentials management, malware detection,
software patches distribution, and trustworthy communi- IV. STATE OF THE ART
cations, to detect, validate, and countermeasure attacks. The state of the art of edge computing in 5G networks
The advantage is that, due to the close proximity of edge is presented according to the three categories, namely fog-
computing, malicious entities can be quickly detected based, MEC-based, and hybrid solutions. Table 2 presents a
and isolated, and real-time responses can be initiated summary of qualitative comparison, which has been used in
to ameliorate the effects of the attacks. This helps to the literature [63], among the existing schemes, covering their
minimize service disruptions. In addition, the scalabil- objectives, computational platforms, attributes, performance
ity and modularity nature, as well as the capabilities, measures, and roles.
of edge computing can facilitate the deployment of block
chain [64] among UEs with limited capabilities. A. FOG BASED SOLUTIONS
In [65], a cross-layer resource management scheme is pre-
F. PERFORMANCE MEASURES sented between optical network and fog computing over
There are three main performance measures as follows: fiber networks in order to incorporate delay requirements
P.1 Lower operational cost. Edge computing reduces the to edge servers. The proposed scheme achieves the objec-
operational cost by providing local functions (R.1)-(R.5), tives of improving QoS (O.2) and improving resource man-
instead of offloading (or sending) tasks and data to the agement (O.5) by providing local computation (R.2). The
cloud. This reduces offloading overhead (or reduces net- proposed scheme uses the hybrid computational platform
work resource consumption), such as bandwidth. (C.3), whereby the edge clouds perform real-time tasks

127282 VOLUME 7, 2019

N. Hassan et al.: Edge Computing in 5G: Review

and the cloud servers perform highly computational and super extreme node. The proposed architecture has shown
resource-intensive tasks. 5G function, including SDN (U.1), to provide lower operational cost (P.1) and higher QoS (P.2)
is used. The proposed scheme has the attribute of low latency (i.e., lower decision making delay).
and close proximity (T.1). The services for different appli-
cations are performed in three layers (i.e., the cloud, fog, and B. MEC BASED SOLUTIONS
UE layers). Highly computational and resource-intensive ser- In [68], an architecture is presented to perform energy-aware
vices are executed at the centralized cloud layer and real-time offloading, whereby each mobile UE decides whether to per-
services are offloaded to the fog layer; while the UE layer form or offload computational tasks to MEC server, in order
performs functions locally at the UEs (or end devices), which to reduce energy consumption of MEC. The UEs are het-
have less computational power and storage due to their lim- erogeneous in nature as they have different communication
itations. The proposed scheme has shown to provide higher and computing capabilities, and the energy consumption of
QoS (P.2) (i.e., lower end-to-end delay). the computational tasks at the mobile UEs is higher than that
In [66], a comparison of energy efficiency between cloud in the MEC server [69]. The proposed architecture achieves
computing and fog computing is made under different mod- the objectives of improving data management (O.1) and
ulation schemes, including 64 quadrature amplitude mod- improving QoS (O.2) by providing local computation (R.2)
ulation (QAM), 16 QAM, phase shift keying (PSK), and and local decision making (R.4). The proposed architecture
quadrature phase shift keying (QPSK), in 5G in order to pro- uses edge computing (C.2) to perform energy-aware offload-
pose an energy-efficient model to improve average through- ing. 5G function, including dynamic access to RATs (U.4),
put and energy consumption per user. The proposed scheme is used. The proposed architecture has the attribute of low
achieves the objectives of improving QoS (O.2) by provid- latency and close proximity (T.1). There are three main steps.
ing local computation (R.2) and local operations (R.5). The Firstly, mobile UEs are classified according to their energy
proposed scheme uses edge computing (C.2), particularly consumption in computation and file transmission, and the
fog computing, to analyze fog computing and its energy transmission delay between the mobile UEs and the MEC.
consumption as compared to cloud computing. 5G function, There are three main categories: a) type 1 UEs use MEC
including dynamic access to RATs (U.4), is used. The pro- server for computation; b) type 2 UEs perform computation
posed scheme has the attribute of low latency and close themselves; and c) type 3 UEs can choose to perform com-
proximity (T.1). Different RATs (i.e., 3G, 4G, and 5G) serve putation either at MEC servers or by themselves. Secondly,
a number of different UEs. An energy efficiency model is priorities are given to the different UEs based on their energy
proposed based on throughput, energy consumption, and the consumption, as well as available channels and their channel
energy consumption level under fog environment. The pro- quality. In general, type 1 UEs enjoy higher priorities due
posed scheme has shown to improve energy efficiency (P.3). to their limited computational capabilities and the need to
In [67], an architecture that enables edge servers to provide offload computational tasks to the MEC server in order to
caching, computing, and communications functions (also satisfy the delay constraint. Thirdly, channels are allocated for
known as 3Cs) is proposed so that content and service UEs with different priorities. Since UEs with higher priorities
providers can deploy their functions, services, and contents are offloaded, there are lower number of UEs competing for
closed to UEs. The proposed architecture achieves the objec- channels. The proposed architecture has shown to provide
tives of improving QoS (O.2) and improving resource man- higher energy efficiency (P.3).
agement (O.5) by providing local storage (R.1) and local In [12], MEC services are autonomously created by the
computation (R.2). The proposed architecture uses edge com- nearest edge server in order to provide mobile UEs with
puting (C.2), particularly fog computing, to reduce process- seamless QoE in video streaming. The proposed scheme
ing delay. 5G functions, including SDN (U.1) and NFV (U.2), achieves the objectives of improving QoS (O.2) and predict-
are used. The proposed architecture has the attributes of ing network demand (O.3) by providing local storage (R.1),
low latency and close proximity (T.1) and network context local computation (R.2), and local decision making (R.4).
awareness (T.3) to acquire network information and traf- The proposed scheme uses edge computing (C.2) to perform
fic distribution. The architecture consists of: a) virtual fog uninterrupted video streaming. 5G function, including D2D
(vFog) which is a framework that empowers UEs with 3Cs communication (U.5), is used. The proposed scheme has the
using NFV so that service provisioning becomes flexible; attributes of low latency and close proximity (T.1), location
b) hyper fog which is a constellation of vFogs that allows data awareness (T.2), and network context awareness (T.3). The
exchange and processing among the vFogs in order to provide edge server receives all or part of a content (e.g., video) from
resources from more than a single vFog; c) regular extreme the cloud, so that the content can be transmitted to UEs with
node, which is a UE with processing and communication reduced delay. Hence, the quality of the content is good as
capabilities; and d) super extreme node is a UE with 3Cs long as a UE is in the vicinity of the edge server. In general,
that manage and manipulate the edge node of vFogs. In this the UEs receive contents from the edge server to reduce delay
architecture, a regular extreme node informs its correspond- (and hence, higher quality streaming); however, if the con-
ing super extreme node about its available resources, and tents are unavailable in the edge server, the UEs would receive
then receive and execute networking tasks assigned by the contents from the cloud, which increases delay (and hence,

VOLUME 7, 2019 127283

 N. Hassan et al.: Edge Computing in 5G: Review

lower quality streaming). There are two main mechanisms In [71], an application-aware traffic redirection mechanism
to ensure seamless content transmission. Firstly, migration is proposed for MEC in order to reduce response time and
enables seamless content transmission when a UE moves bandwidth consumption. The proposed scheme achieves the
from the vicinity of an edge server to another. Secondly, objectives of improving data management (O.1) and improv-
handover enables seamless content transmission when a UE ing QoS (O.2) by providing local computation (R.2), local
handover from a network provider to another, which reduces decision making (R.4), and local operation (R.5). The pro-
delay (and hence, higher quality streaming). The proposed posed scheme uses edge computing (C.2). 5G functions,
scheme has shown to provide higher QoS (P.2) (i.e., lower including dynamic access to RATs (U.4) and D2D communi-
end-to-end delay). cation (U.5), are used. The proposed scheme has the attributes
In [52], a D2D architecture is proposed for a mas- of low latency and close proximity (T.1) and network context
sive number of UEs to execute collaborative tasks in an awareness (T.3) to acquire network information and traffic
energy-efficient manner. The proposed architecture achieves distribution (T.3). The MEC controller allows UEs to offload
the objectives of improving QoS (O.2) and improving (or redirect) the traffic of an application to MEC at the
resource management (O.5) by providing local computation edge of the network when the bandwidth requirement of the
(R.2), local decision making (R.4), and local operation (R.5). traffic exceeds a preset threshold. Subsequently, the UEs can
The proposed architecture uses edge computing (C.2) to per- access the application and its traffic. The proposed scheme
form energy-efficient task offloading. 5G function, including has shown to provide lower operational cost (P.1) and higher
D2D communication (U.5), is used. The proposed archi- QoS (P.2) (i.e., lower response time).
tecture has the attributes of low latency and close proxim- In [72], a virtualized multi-access edge computing frame-
ity (T.1), and network context awareness (T.3) to acquire work is proposed to increase available bandwidth and reduce
network information and traffic distribution. The UEs are end-to-end delay in an intelligent manner in Internet of things.
categorized based on their computational capacity and links The proposed framework achieves the objectives of improv-
(i.e., cellular and D2D links). The UEs use graph matching ing data management (O.1) and improving QoS (O.2) by
to determine whether to perform tasks locally or to offload providing local computation (R.2), and local decision making
them to the edge nodes via D2D in order to achieve energy (R.4). The proposed framework uses edge computing (C.2).
efficiency. The graph matching algorithm, which represents 5G function, including NFV (U.2), is used. The proposed
nodes and links in a graph, has two main stages: a) prunes out framework has the attributes of low latency and close prox-
nodes without tasks; and b) creates a sub-graph that consists imity (T.1), and network context awareness (T.3) to acquire
of nodes with tasks (also called task nodes), replicates them, network information and traffic distribution. The proposed
and connects them to edge nodes of the graph. A task node framework uses MEC to perform virtualized multi-access
performs a task locally if it can be matched by its own replica, computing at the edge of the network. Hardware devices
and it offloads the task to another node if it can be matched are disaggregated and virtualized into layers that provide
with the other node. The proposed architecture has shown to different control functions (e.g., traffic offloading), services
provide higher energy efficiency (P.3). (e.g., computational and storage capabilities), and resources
In [70], a predictive and proactive caching approach is (e.g., computing and storage resources) using NFV. In addi-
introduced in order to reduce peak traffic demands. The tion, traffic offloading provides network traffic information,
proposed scheme achieves the objectives of improving data such as the number of packets, as well as the priority level
management (O.1) and predicting network demand (O.3) by and type of traffic, based on data flow. Traffic is prioritized
providing local storage (R.1) and local computation (R.2). and segregated into three categories, namely high-, medium-,
The proposed scheme uses edge computing (C.2) to perform and low-priority traffic, based on the packet flow rate and
proactive caching at the edge of the network or at UEs. type of traffic, as well as the number of packets in the queue.
5G function, including D2D communication (U.5), is used. Low-priority packets are dropped when signal strength is low
The proposed scheme has the attributes of low latency and and congestion occurs. The proposed framework has shown
close proximity (T.1), and network context awareness (T.3) to to provide lower operational cost (P.1) and higher QoS (P.2)
acquire network information and traffic distribution. Popular (i.e., lower end-to-end delay).
contents are cached in edge servers, BSs, or UEs during off In [73], a fiber wireless (FiWi) access architecture is
peak times.WThe popularity of a content is based on the introduced to improve MEC services (e.g., traffic and net-
UEs’ behavior and the frequency of the BS requesting for work performance monitoring). The proposed architecture
the content. When a BS requests for a particular content, achieves the objectives of improving resource management
there are two possibilities: a) the content is available at an (O.5) by providing local computation (R.2), local deci-
influential UE, who had possessed or processed the content sion making (R.4), and local operation (R.5). The proposed
in the past, and so the content is delivered from the influential scheme uses edge computing (C.2). 5G functions, including
UE to the BS via D2D; and b) the content is unavailable at dynamic access to RATs (U.4) and D2D communication
any influential UEs, and so the content is delivered from the (U.5), are used. The proposed architecture has the attributes
core network to the BS. The proposed scheme has shown to of low latency and close proximity (T.1) and network con-
provide lower operational cost (P.1). text awareness (T.3) to acquire network information and

127284 VOLUME 7, 2019

N. Hassan et al.: Edge Computing in 5G: Review

traffic distribution. In the proposed architecture, BSs serve access to RATs (U.4), are used. The proposed architecture
as rational service centers that provide updated informa- has the attribute of low latency and close proximity (T.1).
tion (i.e., traffic demand and RAT) to backhaul in order to MEC is used to provide a flexible architecture, whereby the
provide intelligent and energy-efficient schemes. MEC is user plane, which is the bottom layer, consists of UEs that
operating over FiWi [74], and ethernet is used to transfer can be grouped into virtual groups (or clusters) based on
traffic from RAN. The FiWi, along with ethernet, provides their ownership, as well as co-location and co-service rela-
a framework for back-haul and broadband access. The pro- tionships that define the relative location between a cluster
posed architecture has shown to provide higher QoS (P.2) and a service requested by the UEs. MEC is deployed close
(i.e., lower queuing delay in the data buffer) and lower energy to the UEs, and the flexibility of the architecture allows
consumption (P.3). the location and structure of the MEC to be customized
In [75], a group of vehicular neighboring nodes (or VNG) and redefined as time goes by. The computational resources
is dynamically managed using SDN to improve control over (e.g., servers, processors, and cloud), as well as radio inter-
network and its resources in vehicular networks. The pro- faces and schemes (e.g., modulation schemes and TDMA)
posed scheme achieves the objectives of improving data are distributed in different slices in network slicing, which
management (O.1), improving QoS (O.2), and improving enables virtualization by running multiple logical networks
resource management (O.5) by providing local computation on a shared physical network infrastructure. The key benefit
(R.2), local decision making (R.4), and local operation (R.5). of network slicing is that it provides an end-to-end virtual
The proposed scheme uses edge computing (C.2). 5G func- network encompassing networking, computation, and storage
tion, including SDN (U.1), is used. The proposed architec- functions. Urgent services (e.g., mission critical services)
ture has the attributes of low latency and close proximity are executed in higher priority slices (e.g., real-time services
(T.1), and network context information and traffic distribu- such as life-saving services in e-health). Hence, additional
tion (T.3). The proposed scheme integrates SDN to MEC in resources are allocated to higher priority slices to serve the
order to strengthen network control (e.g., a unified network urgent services. The proposed architecture has shown to pro-
control of heterogeneous networks) at the edge of the network vide higher QoS (P.2) (i.e., lower end-to-end delay).
for achieving a flexible network control and management.
Real-time instructions (e.g., safety messages) is passed from C. HYBRID SOLUTIONS
road side units to vehicles in order to monitor network states In [77], a D2D-based mobile edge and fog computing archi-
(i.e., the available resources of vehicles) in order to make tecture is introduced to enable collaborative computing,
effective decisions (i.e., road blocks and route changes). which performs tasks in more than a single computing plat-
Using SDN, the edge of the network is segregated into three forms or paradigms, in order to enhance MEC. The pro-
layers: a) the control plane enables the MEC to obtain the posed architecture achieves the objectives of improving data
global knowledge of network states for making optimal deci- management (O.1) and improving QoS (O.2) by providing
sions (i.e., network-level decisions for efficient networking local storage (R.1), local computation (R.2), local data anal-
and fault diagnosis) with lower response time; b) the social ysis (R.3), and local decision making (R.4). The proposed
plane, which is abstracted for communication among VNGs, architecture uses hybrid computing (C.3) to exploit D2D
enables the SDN switch to separate and forward sociality communication in collaborative environment. 5G function,
flows, each of which consists of data packets that indicate the including D2D communication (U.5), is used. The proposed
key features of a VNG (e.g., the strength of a relationship, architecture has the attribute of low latency and close prox-
contact time, contact frequency, and the contact method) imity (T.1). Each UE initiates a service request and send it to
among vehicles so that suitable vehicles can be selected to the nearest relay gateway, which has connection to the core
form strong and weak ties. As an example, two workmates network (or cloud). The service handler of a relay gateway,
from the same office leaving a parking area on a daily basis which has information about the available services, decides
can form a strong tie. As another example, random vehicles whether the requested service should be performed locally
on the road can form temporary weak ties; and c) the data or forwarded to another relay gateway that can perform the
plane provides data transmission. The proposed scheme has service. The decision is based on the availability of the service
shown to provide higher QoS (P.2) (i.e., lower end-to-end (e.g., the processing, computational, and storage capabilities,
delay). as well as delay requirements) at the relay gateway and its
In [76], a non-standalone (i.e., disconnected from the Inter- neighboring gateways. The proposed architecture has shown
net) MEC-based architecture is presented for mission-critical to provide lower operational cost (P.1) and higher QoS (P.2)
public safety services in order to achieve the delay require- (i.e., lower end-to-end and round-trip delays).
ment (i.e., less than 1 ms (ideal) or 10 ms (maximum) of In [78], a context-aware, real-time collaborative architec-
round trip time) of 5G. The proposed architecture achieves the ture is proposed to manage heterogeneous resources (e.g., dif-
objective of improving QoS (O.2) by providing local compu- ferent storage and computational capabilities in different
tation (R.2), local decision making (R.4), and local operation computational platforms/ layers) at the edge of the net-
(R.5). The proposed architecture uses edge computing (C.2). work. The proposed architecture achieves the objective of
5G functions, including SDN (U.1), NFV (U.2), and dynamic improving resource management (O.5) by providing local

VOLUME 7, 2019 127285

 N. Hassan et al.: Edge Computing in 5G: Review

computation (R.2), local decision making (R.4), and local computational approaches, such as hybrid computing (C.3),
operation (R.5). The proposed architecture uses hybrid com- can be used. Edge computing can be used to maintain network
puting (C.3) to optimally distribute tasks among cloud, or service states (e.g. the availability and cost of the links,
MEC, and mobile UEs. 5G function, including dynamic as well as the way a switch forwards the traffic) for evolv-
access to RATs (U.4), is used. The proposed architecture ing applications (e.g., 4K video streaming [80]) and offer
has the attributes of low latency and close proximity (T.1) proxying functionality on behalf of UEs. By maintaining
and network context awareness (T.3). Tasks are split and the network states, the trade-off between the availability and
offloaded among cloud, MEC, and mobile UEs based on QoE performance can be achieved with reduced signaling
the task requirements: a) UEs process tasks that require less overhead incurred by network processes (e.g., handover).
processing and computational capabilities; b) MEC server The signaling messages can also be aggregated to reduce
processes delay-sensitive tasks; and c) cloud processes non- signaling overhead. This leads to reduced network conges-
delay-sensitive tasks. Both MEC and cloud process tasks that tion, hence improving network scalability and network per-
require higher processing and computational capabilities. The formance (e.g., higher throughput [81]). Addressing this open
proposed architecture has shown to provide lower operational issue can provide improvement in QoS (P.2).
cost (P.1) and higher QoS (P.2).
In [17], a real-time, context-aware, service-composition, B. STANDARDIZATION OF PROTOCOLS
and collaborative architecture is proposed to deliver fast Standardization of protocols requires standardizing bodies or
composite service, which is the consolidation of multiple organizations to provide a set of universally acceptable rules
services supported by the collaboration of different hardware for edge computing in 5G environment. There are two main
(e.g., UEs, edge clouds, and cloud) and software with dif- challenges. Firstly, it is difficult to agree upon a standard
ferent capabilities. The proposed architecture achieves the (e.g., the location and capabilities of the edge cloud) due
objectives of improving data management (O.1) and improv- to its flexibility and diversified customization by different
ing QoS (O.2) by providing local computation (R.2), local vendors. Secondly, a large number of heterogeneous UEs
data analysis (R.3), local decision making (R.4), and local use different interfaces to communicate with the edge cloud.
operation (R.5). The proposed architecture uses hybrid com- Standardization effort, such as the initiative from the Euro-
puting (C.3) that enables collaboration among cloud, MEC, pean Telecommunications Standards Institute (ETSI) [82],
and UEs. 5G function, including dynamic access to RATs has been put in place so that heterogeneous UEs can com-
(U.4), is used. The proposed architecture has the attributes municate with edge servers, and different layers and com-
of low latency and close proximity (T.1) and network context putation paradigms can collaborate among themselves, in a
awareness (T.3). Frequently accessed blocks, which are small multi-vendor environment.
units decomposed from a file, are stored (or cached) in MEC
servers. Blocks requested by more than a single server are C. ADDRESSING HETEROGENEITY
replicated and cached in other MEC servers based on file Heterogeneity in communication (e.g., transmission range
types and contents. This helps to reduce the end-to-end delay and data rate) and computing (e.g., hardware architecture
incurred to access the cloud. The proposed architecture has and operating systems) technologies in edge computing
shown to provide lower operational cost (P.1) and higher for 5G has resulted in difficulties in developing a solu-
QoS (P.2). tion that is portable across different environment. Software-
based (or programming-based) schemes may develop a
V. OPEN RESEARCH ISSUES programming-model for edge nodes to facilitate the execution
This section highlights the open research issues for a success- of workloads simultaneously at multiple hardware levels [2].
ful deployment of edge cloud in the 5G environment. However, a comprehensive distributed computing system
must allow the different schemes to operate in a collaborative
A. SERVICE ENHANCEMENT: QOE manner. Data and task-level parallelism splits workload into
QoE is a measure of the overall customer satisfaction level independent and smaller tasks that can be executed in parallel
with a service provider. QoE is related to, but differs from, across different hardware and layers in edge clouds [12]. The
QoS, which embodies the notion that the hardware char- proposed solutions enable heterogeneous UEs to communi-
acteristics (e.g., the storage capacity and the number of cate with edge servers.
processors in the servers [79]) and software characteristics
(e.g., the interface development) can be measured, improved, D. SECURITY AND PRIVACY
and guaranteed. The challenge is to achieve a balanced While security and privacy is enhanced in edge comput-
trade-off between: a) higher availability or seamless connec- ing as data do not travel across a network, there are two
tivity of an application, which can be provided by the cloud main problems that can increase network vulnerability at the
when a UE is out of the vicinity of the edge server; and edge of network. Firstly, the dynamic environment causes
b) higher QoE of the application, which can be provided by the data and network requirements of different network
the edge cloud when the UEs are in the vicinity of the edge entities to vary rapidly. Secondly, the increasing number
server, in order to reduce delay and jitter. Hence, collaborative of devices communicating with each other must require a

127286 VOLUME 7, 2019

N. Hassan et al.: Edge Computing in 5G: Review

scalable solution. Hence, trust and security management must [9] A. Ahmed and E. Ahmed, ‘‘A survey on mobile edge computing,’’ in Proc.
address the aforementioned problems in order to address net- 10th Int. Conf. Intell. Syst. Control (ISCO), Jan. 2016, pp. 1–9.
[10] Q. V. Pham , F. Fang, V. N. Ha, M. Le, Z. Ding, L. B. Le, and
work vulnerability; however, this may incur high complexity W. J. Hwang, ‘‘A survey of multi-access edge computing in 5G and
and cost. Enhancing security and privacy is significant due to beyond: Fundamentals, technology integration, and state-of-the-art,’’ 2019,
the importance of the data (e.g., health information). There arXiv:1906.08452. [Online]. Available: https://arxiv.org/abs/1906.08452
[11] K. Antonakoglou, X. Xu, E. Steinbach, T. Mahmoodi, and M. Dohler,
are two potential solutions. Firstly, applications running on ‘‘Toward haptic communications over the 5G tactile Internet,’’
edge cloud must be blind/ unaware to the raw information IEEE Commun. Surveys Tuts., vol. 20, no. 4, pp. 3034–3059, 4th Quart.,
(or unprocessed data). So, the raw information (e.g., personal 2018.
[12] T. Taleb, S. Dutta, A. Ksentini, M. Iqbal, and H. Flinck, ‘‘Mobile edge
data including healthcare information) must be encrypted or computing potential in making cities smarter,’’ IEEE Commun. Mag.,
processed. Secondly, raw information can be removed prior vol. 55, no. 3, pp. 38–43, Mar. 2017.
to reaching the edge cloud to ensure privacy [83], [84]. [13] Y. Yu, ‘‘Mobile edge computing towards 5G: Vision, recent progress, and
open challenges,’’ China Commun., vol. 13, no. Supplement2, pp. 89–99,
2016.
VI. CONCLUSION [14] S. Raza, S. Wang, M. Ahmed, and M. R. Anwar, ‘‘A survey on vehicular
In this paper, we present a review of the state-of-the edge computing: Architecture, applications, technical issues, and future
art development in edge computing, including fog-based, directions,’’ Wireless Commun. Mobile Comput., vol. 2019, Feb. 2019,
Art. no. 3159762.
MEC-based, and hybrid solutions, in 5G networks. A tax- [15] S. York and R. Poynter, ‘‘Global mobile market research in 2017,’’
onomy is established in which the edge computing in Mobile Research. New York, NY, USA: Springer, Aug. 2017,
approaches are classified according to different characteris- pp. 1–14.
[16] Z. Cui and J. Wang, ‘‘Enhanced software-defined network controller
tics (e.g, objectives, computational platforms, and attributes) to support ad-hoc radio access networks,’’ U.S. Patent 15 054 180,
and the features of edge computing are presented. The key Oct. 23, 2018.
requirements of edge computing are to provide real-time [17] I. A. Ridhawi, M. Aloqaily, Y. Kotb, Y. A. Ridhawi, and Y. Jararweh,
‘‘A collaborative mobile edge computing and user solution for service
interaction, local processing, high data rate, and high avail- composition in 5G systems,’’ Trans. Emerg. Telecommun. Technol., vol. 29,
ability. Edge computing improves network performance no. 11, Jun. 2018, Art. no. e3446.
to support and deploy different scenarios, such as remote [18] B. Yang, Z. Yu, J. Lan, R. Zhang, J. Zhou, and W. Hong, ‘‘Digital
beamforming-based massive MIMO transceiver for 5G millimeter-wave
surgery. Open issues for the successful deployment of edge
communications,’’ IEEE Trans. Microw. Theory Techn., vol. 66, no. 7,
computing in 5G are identified, including service enhance- pp. 3403–3418, Jul. 2018.
ment, standardization, as well as addressing heterogeneity [19] N. C. Luong, P. Wang, D. Niyato, Y.-C. Liang, Z. Han, and F. Hou,
and security vulnerabilities. A qualitative comparison among ‘‘Applications of economic and pricing models for resource management
in 5G wireless networks: A survey,’’ IEEE Commun. Surveys Tuts., to be
the existing schemes in the literature is presented, and it published.
shows the research gaps in this topic whereby the missing [20] Z. Pi, J. Choi, and R. W. Heath, Jr., ‘‘Millimeter-wave gigabit broadband
ticks represent potential open issues that can be further evolution toward 5G: Fixed access and backhaul,’’ IEEE Commun. Mag.,
vol. 54, no. 4, pp. 138–144, Apr. 2016.
investigated. Although the deployment of edge computing [21] A. Ateya, A. Muthanna, I. Gudkova, A. Abuarqoub, A. Vybornova, and
in 5G provides numerous benefits, the convergence of both A. Koucheryavy, ‘‘Development of intelligent core network for tactile
edge computing and 5G brings about new issues that should Internet and future smart systems,’’ J. Sens. Actuator Netw., vol. 7, p. 1,
Jan. 2018.
be resolved in the near future. [22] Z. Ning, X. Kong, F. Xia, W. Hou, and X. Wang, ‘‘Green and sustainable
cloud of things: Enabling collaborative edge computing,’’ IEEE Commun.
REFERENCES Mag., vol. 57, no. 1, pp. 72–78, Jan. 2018.
[23] K. Yeow, A. Gani, R. W. Ahmad, J. Rodrigues, and K. Ko, ‘‘Decentralized
[1] W. Z. Khan, E. Ahmed, S. Hakak, I. Yaqoob, and A. Ahmed, ‘‘Edge
consensus for edge-centric Internet of Things: A review, taxonomy, and
computing: A survey,’’ Future Gener. Comput. Syst., vol. 97, pp. 219–235,
research issues,’’ IEEE Access, vol. 6, pp. 1513–1524, 2017.
Aug. 2019.
[2] N. Hassan, S. Gillani, E. Ahmed, I. Ibrar, and M. Imran, ‘‘The role of edge [24] M. Simsek, A. Aijaz, M. Dohler, J. Sachs, and G. Fettweis, ‘‘5G-enabled
computing in Internet of Things,’’ IEEE Commun. Mag., vol. 56, no. 11, tactile Internet,’’ IEEE J. Sel. Areas Commun., vol. 34, no. 3, pp. 460–473,
pp. 110–115, Nov. 2018. Mar. 2016.
[3] E. Ahmed, A. Ahmed, I. Yaqoob, J. Shuja, A. Gani, M. Imran, and [25] S. Choy, B. Wong, G. Simon, and C. Rosenberg, ‘‘The brewing storm
M. Shoaib, ‘‘Bringing computation closer toward the user network: Is in cloud gaming: A measurement study on cloud to end-user latency,’’
edge computing the solution?’’ IEEE Commun. Mag., vol. 55, no. 11, in Proc. 11th Annu. Workshop Netw. Syst. Support Games (NetGames),
pp. 138–144, Nov. 2017. Nov. 2012, pp. 1–6.
[4] J. Shuja, S. Mustafa, R. W. Ahmad, S. A. Madani, A. Gani, and M. K. Khan, [26] Z. Zhao, K. Hwang, and J. Villeta, ‘‘Game cloud design with virtualized
‘‘Analysis of vector code offloading framework in heterogeneous cloud and CPU/GPU servers and initial performance results,’’ in Proc. 3rd Workshop
edge architectures,’’ IEEE Access, vol. 5, pp. 24542–24554, 2017. Sci. Cloud Comput. Date ScienceCloud, 2012, pp. 23–30.
[5] P. Popovski, K. F. Trillingsgaard, O. Simeone, and G. Durisi, ‘‘5G wire- [27] W. Shi and S. Dustdar, ‘‘The promise of edge computing,’’ Computer,
less network slicing for eMBB, URLLC, and mMTC: A communication- vol. 49, no. 5, pp. 78–81, 2016.
theoretic view,’’ IEEE Access, vol. 6, pp. 55765–55779, 2018. [28] W. Hu, Y. Gao, K. Ha, J. Wang, B. Amos, Z. Chen, and P. Pillai, ‘‘Quanti-
[6] A. Anand, G. D. Veciana, and S. Shakkottai, ‘‘Joint scheduling of fying the impact of edge computing on mobile applications,’’ in Proc. 7th
URLLC and eMBB traffic in 5G wireless networks,’’ in Proc. INFOCOM, ACM SIGOPS Asia–Pacific Workshop Syst., 2016, pp. 1–5.
Apr. 2018, pp. 1970–1978. [29] M. Fahad, F. Aadil, S. Khan, P. A. Shah, K. Muhammad, J. Lloret, H. Wang,
[7] J. M. Hamamreh, E. Basar, and H. Arslan, ‘‘OFDM-subcarrier index J. W. Lee, and I. Mehmood, ‘‘Grey wolf optimization based clustering
selection for enhancing security and reliability of 5G URLLC services,’’ algorithm for vehicular ad-hoc networks,’’ Comput. Elect. Eng., vol. 70,
IEEE Access, vol. 5, pp. 25863–25875, 2017. pp. 853–870, Aug. 2018.
[8] P. Gallo, K. Kosek-Szott, S. Szott, and I. Tinnirello, ‘‘CADWAN: A con- [30] F. Aadil, A. Raza, M. F. Khan, M. Maqsood, I. Mehmood, and S. Rho,
trol architecture for dense WiFi access networks,’’ IEEE Commun. Mag., ‘‘Energy aware cluster-based routing in flying ad-hoc networks,’’ Sensors,
vol. 56, no. 1, pp. 194–201, Jan. 2018. vol. 18, no. 5, p. 1413, 2018

VOLUME 7, 2019 127287

 N. Hassan et al.: Edge Computing in 5G: Review

[31] P. Porambage, J. Okwuibe, M. Liyanage, M. Ylianttila, and T. Taleb, [53] Y. He, J. Ren, G. Yu, and Y. Cai, ‘‘D2D communications meet mobile edge
‘‘Survey on multi-access edge computing for Internet of Things computing for enhanced computation capacity in cellular networks,’’ IEEE
realization,’’ IEEE Commun. Surveys Tuts., vol. 20, no. 4, pp. 2961–2991, Trans. Wireless Commun., vol. 18, no. 3, pp. 1750–1763, Feb. 2019.
4th Quart., 2018. [54] J. Zhang, W. Xia, F. Yan, and L. Shen, ‘‘Joint computation offloading and
[32] M. Shahid, A. Bo, A. Dulaimi, and K. Tsang, ‘‘Guest editorial 5G tac- resource allocation optimization in heterogeneous networks with mobile
tile Internet: An application for industrial automation,’’ IEEE Trans. Ind. edge computing,’’ IEEE Access, vol. 6, pp. 19324–19337, 2018.
Inform., vol. 15, no. 5, pp. 2992–2994, May 2019. [55] N. Bessis and C. Dobre, Big Data and Internet of Things: A Roadmap for
[33] M. Aazam, K. A. Harras, and S. Zeadally, ‘‘Fog computing for 5G tactile Smart Environments, vol. 546. New York, NY, USA: Springer, 2014.
industrial Internet of Things: QoE-aware resource allocation model,’’ IEEE [56] J. A. Silva, R. Monteiro, H. Paulino, and J. M. Lourenco, ‘‘Ephemeral
Trans. Ind. Informat., vol. 15, no. 5, pp. 3085–3092, May 2019. data storage for networks of hand-held devices,’’in Proc. IEEE Trust-
[34] J. Rao and S. Vrzic, ‘‘Packet duplication for URLLC in 5G: Architectural com/BigDataSE/ISPA, Aug. 2016, pp. 1106–1113.
enhancements and performance analysis,’’ IEEE Netw., vol. 32, no. 2, [57] A. Ndikumana, N. H. Tran, T. M. Ho, Z. Han, W. Saad, D. Niyato, and
pp. 32–40, Mar. 2018. C. S. Hong, ‘‘Joint communication, computation, caching, and control in
[35] R. Miller. (2017). Data Center First: Intels Vision for a Data- big data multi-access edge computing,’’ IEEE Trans. Mobile Comput., to
Driven World. Accessed: Apr. 24, 2019. [Online]. Available: be published.
https://datacenterfrontier.com/data-center-first-intels-vision-for-a-data- [58] C. Li, Y. Xue, J. Wang, W. Zhang, and T. Li, ‘‘Edge-oriented computing
driven-world/ paradigms,’’ ACM Comput. Surv., vol. 51, no. 2, pp. 1–34, Apr. 2018.
[36] X. Sun and N. Ansari, ‘‘Green cloudlet network: A distributed green mobile [59] E. Ahmed, I. Yaqoob, I. A. Hashem, I. Khan, A. I. Ahmed, M. Imran, and
cloud network,’’ IEEE Netw., vol. 31, no. 1, pp. 64–70, Feb. 2017. A. V. Vasilakos, ‘‘The role of big data analytics in Internet of Things,’’
[37] Y. C. Hu, M. Patel, D. Sabella, N. Sprecher, and V. Young, ‘‘Mobile edge Comput. Netw., vol. 129, pp. 459–471, Dec. 2017.
computing—A key technology towards 5G,’’ ETSI White Paper, vol. 11, [60] M. M. Hussain, M. S. Alam, and M. M. S. Beg, ‘‘Feasibility of fog
no. 11, pp. 1–16, Sep. 2015. computing in smart grid architectures,’’ in Proc. 2nd Int. Conf. Commun.,
[38] Z. Zhao, L. Guardalben, M. Karimzadeh, J. Silva, T. Braun, and Comput. Netw. Singapore: Springer, Sep. 2018, pp. 999–1010.
S. Sargento, ‘‘Mobility prediction-assisted over-the-top edge prefetching [61] C. Xu, K. Wang, P. Li, S. Guo, J. Luo, B. Ye, and M. Guo, ‘‘Mak-
for hierarchical VANETs,’’ IEEE J. Sel. Areas Commun., vol. 36, no. 8, ing big data open in edges: A resource-efficient blockchain-based
pp. 1786–1801, Aug. 2018. approach,’’ IEEE Trans. Parallel Distrib. Syst., vol. 30, no. 4, pp. 870–882,
[39] E. Chirivella-Perez J. Gutiérrez-Aguado, J. M. Alcaraz-Calero, and Apr. 2019.
Q. Wang, ‘‘Nfvmon: enabling multioperator flow monitoring in 5G [62] M. Ali, H. S. Bilal, M. A. Razzaq, J. Khan, S. Lee, M. Idris, M. Aazam,
mobile edge computing,’’ Wireless Commun. Mobile Comput., vol. 2018, T. Choi, S. C. Han, and B. H. Kang, ‘‘IoTFLiP: IoT-based flipped learning
Feb. 2018, Art. no. 2860452. platform for medical education,’’ Digital Commun. Netw., vol. 3, no. 3,
pp. 188–194, Aug. 2017.
[40] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, ‘‘Fog computing and its
role in the Internet of Things,’’ in Proc. 1st Ed. MCC Workshop Mobile [63] P. Mach and Z. Becvar, ‘‘Mobile edge computing: A survey on architecture
Cloud Comput., Aug. 2012, pp. 13–16. and computation offloading,’’ IEEE Commun. Surveys Tuts., vol. 19, no. 3,
pp. 1628–1656, Sep. 2017.
[41] General Intelligence, ‘‘Understanding 5G: Perspectives on future tech-
[64] R. Yang, F. R. Yu, P. Si, Z. Yang, and Y. Zhang, ‘‘Integrated blockchain and
nological advancements in mobile,’’ Walbrook Building, London, U.K.,
edge computing systems: A survey, some research issues and challenges,’’
White Paper 26, 2014.
IEEE Commun. Surveys Tuts., vol. 21, no. 2, pp. 1508–1532, Feb. 2019.
[42] Y. Mao, C. You, J. Zhang, K. Huang, and K. B. Letaief, ‘‘A survey on
[65] S. Guo, S. Shao, Y. Wang, and H. Yang, ‘‘Cross stratum resources protec-
mobile edge computing: The communication perspective,’’ IEEE Commun.
tion in fog-computing-based radio over fiber networks for 5G services,’’
Surveys Tuts., vol. 19, no. 4, pp. 2322–2358, 4th Quart., 2017.
Opt. Fiber Technol., vol. 37, pp. 61–68, Sep. 2017.
[43] Y. Li and M. Chen, ‘‘Software-defined network function virtualization:
[66] S. Kitanov and T. Janevski, ‘‘Energy efficiency of fog computing and
A survey,’’ IEEE Access, vol. 3, pp. 2542–2553, 2015.
networking services in 5G networks,’’in Proc. EUROCON, Jul. 2017,
[44] Y. Jararweh, C. Mavromoustakis, D. B. Rawat, and M. H. Rehmani, ‘‘SDN, pp. 491–494.
NFV, and mobile edge computing with QoE support for 5G,’’ Trans. Emerg. [67] E. K. Markakis, K. Karras, A. Sideris, G. Alexiou, and E. Pallis, ‘‘Com-
Telecommun. Technol., vol. 29, no. 11, p. 3536, Nov. 2018. puting, caching, and communication at the edge: The cornerstone for
[45] A. J. Gonzalez, G. Nencioni, A. Kamisiński, B. E. Helvik, and building a versatile 5G ecosystem,’’ IEEE Commun. Mag., vol. 55, no. 11,
P. E. Heegaard, ‘‘Dependability of the NFV orchestrator: State of the art pp. 152–157, Nov. 2017.
and research challenges,’’ IEEE Commun. Surveys Tuts., vol. 20, no. 4, [68] K. Zhang, Y. Mao, S. Leng, Q. Zhao, L. Li, X. Peng, L. Pan, S. Maharjan,
pp. 3307–3329, 4th Quart., 2018. and Y. Zhang, ‘‘Energy-efficient offloading for mobile edge computing in
[46] I. Sarrigiannis, E. Kartsakli, K. Ramantas, A. Antonopoulos, and 5G heterogeneous networks,’’ IEEE Access, vol. 4, pp. 5896–5907, 2016.
C. Verikoukis, ‘‘Application and network VNF migration in a MEC- [69] Y. Ai, M. Peng, and K. Zhang, ‘‘Edge computing technologies for
enabled 5G architecture,’’ in Proc. 23rd IEEE Int. Workshop Comput. Internet of Things: A primer,’’ Digit. Commun. Netw., vol. 4, no. 2,
Aided Modeling Design Commun. Links Netw. (CAMAD), Nov. 2018, pp. 77–86, 2018.
pp. 1–6. [70] E. Bastug, M. Bennis, and M. Debbah, ‘‘Living on the edge: The role of
[47] C. E. Shannon, ‘‘A mathematical theory of communication,’’ Bell Syst. proactive caching in 5G wireless networks,’’ IEEE Commun. Mag., vol. 52,
Tech. J., vol. 27, no. 3, pp. 379–423, 1948. no. 8, pp. 82–89, Aug. 2014.
[48] S. Wang, X. Zhang, Y. Zhang, L. Wang, J. Yang, and W. Wang, ‘‘A survey [71] S.-C. Huang, Y.-C. Luo, B.-L. Chen, Y.-C. Chung, and J. Chou,
on mobile edge networks: Convergence of computing, caching and com- ‘‘Application-aware traffic redirection: A mobile edge computing imple-
munications,’’ IEEE Access, vol. 5, pp. 6757–6779, 2017. mentation toward future 5G networks,’’ in Proc. IEEE 7th Int. Symp. Cloud
[49] S. Parkvall, E. Dahlman, A. Furuskar, and M. Frenne, ‘‘NR: The new 5G Service Comput. (SC2), Nov. 2017, pp. 17–23.
radio access technology,’’ IEEE Commun. Standards Mag., vol. 1, no. 4, [72] H.-C. Hsieh, J.-L. Chen, and A. Benslimane, ‘‘5G virtualized multi-access
pp. 24–30, Dec. 2017. edge computing platform for IoT applications,’’ J. Netw. Comput. Appl.,
[50] Y. He, F. R. Yu, N. Zhao, and H. Yin, ‘‘Secure social networks in vol. 115, pp. 94–102, Aug. 2018.
5G systems with mobile edge computing, caching, and device-to-device [73] B. P. Rimal, D. P. Van, and M. Maier, ‘‘Mobile edge computing empowered
communications,’’ IEEE Wireless Commun., vol. 25, no. 3, pp. 103–109, fiber-wireless access networks in the 5G era,’’ IEEE Commun. Mag.,
Jun. 2018. vol. 55, no. 2, pp. 192–200, Feb. 2017.
[51] H. Wang, J. Wang, G. Ding, L. Wang, T. A. Tsiftsis, and P. K. Sharma, [74] J. Liu, H. Guo, H. Nishiyama, H. Ujikawa, K. Suzuki, and N. Kato,
‘‘Resource allocation for energy harvesting-powered D2D communication ‘‘New perspectives on future smart FiWi networks: Scalability, reliability,
underlaying UAV-assisted networks,’’ IEEE Trans. Green Commun. Netw., and energy efficiency,’’ IEEE Commun. Surveys Tuts., vol. 18, no. 2,
vol. 2, no. 1, pp. 14–24, Mar. 2018. pp. 1045–1072, 2nd Quart., 2016.
[52] X. Chen, L. Pu, L. Gao, W. Wu, and D. Wu, ‘‘Exploiting massive D2D [75] X. Huang, R. Yu, J. Kang, Y. He, and Y. Zhang, ‘‘Exploring mobile edge
collaboration for energy-efficient mobile edge computing,’’ IEEE Wireless computing for 5G-enabled software defined vehicular networks,’’ IEEE
Commun., vol. 24, no. 4, pp. 64–71, Aug. 2017. Wireless Commun., vol. 24, no. 6, pp. 55–63, Dec. 2017.

127288 VOLUME 7, 2019

N. Hassan et al.: Edge Computing in 5G: Review

[76] R. Solozabal, A. Sanchoyerto, E. Atxutegi, B. Blanco, J. O. Fajardo, and KOK-LIM ALVIN YAU (M’08–SM’18) received
F. and Liberal, ‘‘Exploitation of mobile edge computing in 5G distributed the B.Eng. degree (Hons.) in electrical and elec-
mission-critical push-to-talk service deployment,’’ IEEE Access, vol. 6, tronics engineering from Universiti Teknologi
pp. 37665–37675, 2018. PETRONAS, Malaysia, in 2005, the M.Sc. degree
[77] S. Singh, Y.-C. Chiu, Y.-H. Tsai, and J.-S. Yang, ‘‘Mobile edge fog com- in electrical engineering from the National Univer-
puting in 5G era: Architecture and implementation,’’ in Proc. Int. Comput. sity of Singapore, in 2007, and the Ph.D. degree in
Symp. (ICS), Dec. 2016, pp. 731–735. network engineering from the Victoria University
[78] T. X. Tran, A. Hajisami, P. Pandey, and D. Pompili, ‘‘Collaborative mobile
of Wellington, New Zealand, in 2010. He is cur-
edge computing in 5G networks: New paradigms, scenarios, and chal-
rently an Associate Professor with the Department
lenges,’’ IEEE Commun. Mag., vol. 55, no. 4, pp. 54–61, Apr. 2017.
[79] M. Breternitz, K. A. Lowery, P. Kaminski, and A. Chernoff, ‘‘System and of Computing and Information Systems, Sunway
method for allocating a cluster of nodes for a cloud computing system University. He is also a Researcher, a Lecturer, and a Consultant in cog-
based on hardware characteristics,’’ U.S. Patent 13 568 395, Feb. 13, 2014. nitive radio, wireless networks, applied artificial intelligence, applied deep
[80] K. Kanai, K. Imagane, and J. Katto, ‘‘[Invited paper] overview of multi- learning, and reinforcement learning. He serves as a TPC Member and a
media mobile edge computing,’’ ITE Trans. Media Technol. Appl., vol. 6, Reviewer for major international conferences, including ICC, VTC, LCN,
no. 1, pp. 46–52, 2018. GLOBECOM, and AINA. He was a recipient of the 2007 Professional
[81] Z. Guan, Y. Zhang, L. Wu, J. Wu, J. Li, Y. Ma, and J. Hu, ‘‘APPA: Engineer Board of Singapore Gold Medal for being the best graduate of
An anonymous and privacy preserving data aggregation scheme for fog- the M.Sc. degree, in 2006 and 2007. He also served as the Vice General
enhanced IoT,’’ J. Netw. Comput. Appl., vol. 125, pp. 82–92, Jan. 2019. Co-Chair for ICOIN’18, the Co-Chair for IET ICFCNA’14, and the Co-Chair
[82] K. Jain and S. Mohapatra, ‘‘Taxonomy of edge computing: Chal- (Organizing Committee) for IET ICWCA’12. He serves as an Editor for the
lenges, opportunities, and data reduction methods,’’ in Edge Computing. KSII Transactions on Internet and Information Systems, an Associate Editor
New York, NY, USA: Springer, Nov. 2018, pp. 51–69. for IEEE ACCESS, a Guest Editor for the Special Issues of IEEE ACCESS, IET
[83] E. Ahmed and M. H. Rehmani, ‘‘Mobile edge computing: Opportuni- Networks, the IEEE Computational Intelligence Magazine, Springer Journal
ties, solutions, and challenges,’’ Future Generat. Comput. Syst., vol. 70,
of Ambient Intelligence and Humanized Computing, and a Regular Reviewer
pp. 59–63, May 2017.
for more than 20 journals, including the IEEE journals and magazines, the Ad
[84] K. Kaur, S. Garg, G. Kaddoum, M. Guizani, and D. Jayakody,
‘‘A lightweight and privacy-preserving authentication protocol for Hoc Networks, the IET Communications, and others.
mobile edge computing,’’ 2019, arXiv:1907.08896. [Online]. Available:
https://arxiv.org/abs/1907.08896

CELIMUGE WU received the M.E. degree

from the Beijing Institute of Technology, China,
in 2006, and the Ph.D. degree from The Univer-
NAJMUL HASSAN received the M.S. degree in sity of Electro-Communications, Japan, in 2010,
computer sciences from Mohammad Ali Jinnah where he is currently an Associate Professor. His
University, Islamabad, Pakistan, in 2010. He is current research interests include vehicular net-
currently pursuing the Ph.D. degree with the works, sensor networks, intelligent transport sys-
Department of Computing and Information Sys- tems, the IoT, and edge computing. He has been
tems, Sunway University. He was a Visiting a TPC Co-Chair of Wireless Days 2019, ICT-DM
Faculty with the Department of Computer Sci- 2019, ICT-DM 2018, a Track Co-Chair of many
ence, COMSATS University Islamabad, Attock international conferences including IEEE VTC 2020-Spring, ICCCN 2019,
and Abbottabad Campus. His research interests and IEEE PIMRC 2016. He serves as an Associate Editor for IEEE ACCESS,
include the Internet of Things, edge/cloud commu- IEICE Transactions on Communications, the International Journal of Dis-
nication, wireless body area networks, the airborne Internet, and wireless tributed Sensor Networks, and MDPI Sensors.
sensor networks.

VOLUME 7, 2019 127289