SPECIAL SECTION ON EDGE COMPUTING

AND NETWORKING FOR UBIQUITOUS AI

Received April 12, 2020, accepted April 28, 2020, date of publication May 6, 2020, date of current version May 19, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.2991734

An Overview on Edge Computing Research

KEYAN CAO 1,2
, YEFAN LIU 1
, GONGJIE MENG1, AND QIMENG SUN1
1
College of Information and Control Engineering, Shenyang Jianzhu University, Shenyang 110168, China
2
Liaoning Province Big data Management and Analysis Laboratory of Urban Construction, Shenyang 110168, China

Corresponding author: Keyan Cao (caokeyan@sjzu.edu.cn)

This work was supported in part by the National Natural Science Foundation of China under Grant 61602323, in part by the National
Postdoctoral Foundation of China under Grant 2016M591455, in part by the Youth Seedling Foundation of Liaoning Province under Grant
lnqn201913, and in part by the Natural Science Funds of Liaoning Province under Grant 2019MS264 and Grant 20180550019.

ABSTRACT With the rapid development of the Internet of Everything (IoE), the number of smart devices
connected to the Internet is increasing, resulting in large-scale data, which has caused problems such as
bandwidth load, slow response speed, poor security, and poor privacy in traditional cloud computing
models. Traditional cloud computing is no longer sufficient to support the diverse needs of today’s
intelligent society for data processing, so edge computing technologies have emerged. It is a new
computing paradigm for performing calculations at the edge of the network. Unlike cloud computing, it
emphasizes closer to the user and closer to the source of the data. At the edge of the network, it is
lightweight for local, small-scale data storage and processing. This article mainly reviews the related
research and results of edge computing. First, it summarizes the concept of edge computing and compares
it with cloud computing. Then summarize the architecture of edge computing, keyword technology,
security and privacy protection, and finally summarize the applications of edge computing.

INDEX TERMS Edge computing, cloud computing, Internet of Things.

I. INTRODUCTION devices connected to the network will exceed 50
With the development of intelligent society and the billion, and the total global data in 2020 will also
continuous improvement of people’s needs, exceed 40 ZB [2]. Based on the continuous

The associate editor coordinating the review of this

intelligence has involved various industries and manuscript and
people’s daily lives in society. Edge devices have and massive growth of data volume and various
spread to all aspects of society, such as smart data processing requirements, cloud-based big data
homes and autonomous vehicles in the field of processing has shown many shortcomings:
transportation, camera, intelligent production robot Real-time: If a large number of edge devices are
in intelligent manufacturing, etc. As a result, the added, a large amount of terminal data is still
number of devices connected to the Internet has transmitted to the cloud for processing, the
increased significantly. Cisco pointed out in the intermediate data transmission volume will be
Global Cloud Index [1] that in 2016, there were greatly increased, the data transmission
17.1 billion devices connected to the Internet,by performance will be reduced, resulting in a large
2019, the total number of data traffic in global data load of network transmission bandwidth, resulting
centers will reach 10.4 Zettabyte (ZB), 45% of the in data transmission delay. In some application
data will be stored, processed and analyzed on the scenarios that require real-time feedback, such as
edge of the network, and by 2020, the number of traffic, monitoring, etc., cloud computing will not
wireless devices connected to the network will be able to meet business real-time requirements.
exceed 50 billion. The amount of data generated by Security and privacy: For example, when using
devices worldwide has also increased from 218ZB various applications in smartphones, applications
in 2016 to 847 ZB in 2021. International data will require user data, including privacy data. There
company Internet Data Center (IDC) statistics is a high risk of privacy leakage or attack on this
show that by 2020, the number of terminals and data when uploaded to the cloud center.
 K. Cao et al.: Overview on Edge Computing Research

Energy consumption: the number of smart between the data source and the path of cloud
devices continues to increase, and the power computing center.’’ Satyanarayanan, a professor at
consumption of data centers in China has increased Carnegie Mellon university in the United States,
significantly. Improving the use efficiency of cloud describes edge computing as: ‘‘Edge computing is
computing energy consumption [3] cannot meet the a new computing model that deploys computing
increasing demand for data energy consumption. and storage resources (such as cloudlets, micro data

approving it for publication was Sabah Mohammed. computing.

85714 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ VOLUME 8, 2020

The rapidly developing intelligent society will have centers, or fog nodes, etc.) at the edge of the
higher requirements for the energy consumption of network closer to mobile devices or sensors’’ [4].
cloud Due to the increasing amount of data and Zha et al. [8] proposed on the basis of the above
increasing requirements for data processing, edge two definitions: ‘‘Edge computing is a new
computing has emerged at the historic moment. computing model that unifies resources that are
Edge computing technology provides artificial close to the user in geographical distance or
intelligence services for rapidly growing terminal network distance to provide computing, storage,
devices and data, and makes services more stable. and network for applications service.’’ China’s
Edge computing is close to the source of the data, edge computing industry alliance defines edge
such as smart terminals. It stores and processes computing as: ‘‘near the edge of the network or the
data at the edge of the network. It has proximity source of the data, an open platform that integrates
and location awareness, and provides users with core capabilities such as networking, computing,
near-end services. In terms of data processing, it is storage, applications, and provides edge intelligent
faster, real-time, and secure. It can also solve the services nearby to meet the industry agility key
problem of excessive energy consumption in cloud requirements in connection, real-time business,
computing, reduce costs, and reduce the pressure data optimization, application intelligence, security
of network bandwidth. Edge computing is applied and privacy’’ [9].
in various fields such as production, energy, smart In other words, edge computing is to provide
home, and transportation. services and perform calculations at the edge of the
With the development of the Internet of Things network and data generation. Edge computing is to
(IoT), edge computing models are urgently needed migrate the cloud’s network, computing, storage
and have become hot research issues. In this capabilities and resources to the edge of the
article, we introduce edge computing in detail from network, and provide intelligent services at the
the aspects of edge computing introduction, edge to meet the critical needs of the IT industry in
architecture, key technologies, security and agile linking, real-time business, data optimization,
privacy, and applications to provide a reference for application intelligence, security and privacy, and
edge computing researchers. meets the requirements of low latency and high
bandwidth on the network. Edge computing has
become a research hotspot nowadays [10]–[14].
II. EDGE COMPUTING
A. EDGE COMPUTING CONCEPTS
B. CLOUD COMPUTING AND EDGE
Edge computing is different from traditional cloud COMPUTING1) CLOUD COMPUTING
computing. It is a new computing paradigm that
Before the emergence of edge computing,
performs computing at the edge of the network. Its
traditional cloud computing transfers all data to the
core idea is to make computing closer to the source
cloud computing center through the network, and
of the data [4]. Researchers have different
solves the computing and storage problems in a
definitions of edge computing. Shi et al. [5]–[7]
centralized way. In literature [15], the development
introduced
history of cloud computing is described. In the
theemergenceoftheconceptofedgecomputing:‘‘Edge
search engine conference (sessane jose 2006) in
computing is a new computing mode of network
August 2006, the CEO of Google first proposed the
edge execution. The downlink data of edge
concept of cloud computing. In the development
computing represents cloud
history of cloud computing, this is the first time to
service,theuplinkdatarepresentstheInternetofEveryt
formally put forward the concept of cloud
hing,and the edge of edge computing refers to the
computing. With the development of
arbitrary computing and network resources
20179 VOLUME 8, 2020
K. Cao et al. : Overview on Edge Computing Research

searchenginesrepresentedbyGoogle,cloudcomputin manufacturing [21], energy [22], security [23] and

gstarts to show strong vitality. Nowadays, cloud privacy protection [24], and intelligent family. For
computing has gradually developed. It is a very example, in the industrial production of intelligent
powerful network service platform including manufacturing, the role of the cloud is to control
distributed computing, load balancing, parallel the whole. In the edge nodes, it is necessary to have
computing, network storage, virtualization and the function of real-time detection and solve the
other technologies. However, nowadays, with the problems in time. Edge computing takes advantage
popularization and development of the Internet of of the real-time characteristics, and in collaboration
Things in people’s life, the number of devices with cloud computing and synergy, not only
connected to the Internet of Things is gradually improves production efficiency, but also can detect
increasing, and a large amount of data is generated. abnormalities of equipment in a timely manner. In
The network bandwidth of cloud computing has the field of smart home, edge computing nodes
been unable to meet the needs of time-sensitive mainly involve some intelligent terminals. Edge
systems [16] and real-time performance. Therefore, computing nodes calculate heterogeneous data
cloud computing model has great defects in load, from different devices and upload it to the cloud for
real-time [17]–[19], transmission processing, so as to realize the control of edge
bandwidth,energyconsumptionanddatasecurityandp nodes from the cloud and the access of edge nodes
rivacy protection [20]. to the cloud. In order to meet the needs of Internet
of Things devices, cloud computing and edge
2) CONNECTION AND DIFFERENCE BETWEEN CLOUD computing play their respective advantages, and
COMPUTING AND EDGE COMPUTING only the joint development of the two can
continuously promote the progress of the Internet.
The emergence of edge computing will not replace
Edge computing is an extension of cloud
cloud computing. In the aspects of network,
computing, which has its own characteristics with
business, application and intelligence, the two
cloud computing. The main feature of cloud
should exist together, complement each other and
computing are that it can grasp the whole, can
develop in a coordinated way, which will help the
process a large amount of data, conduct in-depth
digital transformation of the industry to a greater
analysis, and also plays an important role in non-
extent. All data onto edge nodes still need to be
real-time data processing, such as business
summarized in the cloud to achieve in-depth
decision-making and other fields. Edge computing
analysis and obtain more meaningful analysis
focuses on the local, and can play a better role in
results. Therefore, cloud computing is still playing
small-scale,real-
an important role in the development of Internet of
timeintelligentanalysis,suchasmeetingthe real-time
Things devices that are gradually intelligent.
needs of local businesses. Therefore, in intelligent
In the context of the Internet of Things, if all the
applications, cloud computing is more suitable for
large amount of data generated by the connected
centralized processing of large-scale data, while
devices are transmitted to the cloud, cloud
edge computing can be used for small-scale
computing will cause a large load. At this time,
intelligent analysis and local services. In terms of
edge computing is required to share the pressure of
network resources, edge computing is responsible
the cloud and take charge of tasks within its scope
for data closer to the information source. Therefore,
of the edge. When there is a problem in edge
data can be stored and processed locally without
computing, the data in the cloud is not lost. In some
uploading all the data to the cloud. The reduction
Internet services, some data needs to be returned to
of network burden greatly improves the utilization
the cloud for processing after being processed by
efficiency of network bandwidth. Cloud computing
edge computing, such as in-depth analysis of data
and edge computing play an important role in the
mining and sharing, which requires the cooperation
future development of intelligent Internet of Things
of cloud computing and edge computing. Both
[25]. The main differences between cloud
developments bring stability to connected devices
computing and edge computing are shown in Table
in the Internet of Things network. The working
1.
method of the two can be that cloud computing is
based on big data analysis and output, passed to the
edge side, and then processed and executed by C. ADVANTAGES OF EDGE COMPUTING
edge computing. Nowadays, the coordinated
Edge computing model stores and processes data
development of the two has been applied in many
on edge devices without uploading to cloud
aspects of real life, such as intelligent
computing platform.
VOLUME 8, 2020 20180
 K. Cao et al.: Overview on Edge Computing Research

TABLE 1. Main differences between cloud computing and edge datatransmittedonthenetwork,reducesthetransmissio

computing.
ncost and network bandwidth pressure, reduces the
energy consumption of local equipment, and
improves the computing efficiency.

III. ARCHITECTURE OF EDGE COMPUTING

With the Internet of Everything era and the
development of 5G, edge computing is considered
as one of the key technologies in the next
Due to this feature, edge computing has obvious generation of communication network following
advantages in the following aspects: the Internet of Things and artificial intelligence
Fast data processing and analysis, real-time: the [25]. The reference architecture for edge computing
rapid growth of data volume and the pressure of is the focus of many organizations. This section
network bandwidth are disadvantages of cloud begins with an overview of the general architecture
computing [26]. Compared with traditional cloud for edge computing, followed by a detailed
computing, edge computing has advantages in introduction of the reference architecture proposed
response speed and real-time. Edge computing is by the edge computing industry alliance (ECC) and
closer to the data source, data storage and the Linux foundation in sections 3.2 and 3.3,
computing tasks can be carried out in the edge respectively.
computing node, which reduces the intermediate
data transmission process. It emphasizes proximity A. GENERAL ARCHITECTURE OF EDGE COMPUTING
to users and provides users with better intelligent Edge computing architecture is a federated network
services, thus improving data transmission structure that extends cloud services to the edge of
performance, ensuring real-time processing and the network by introducing edge devices between
reducing delay time. Edge computing provides terminal devices and cloud computing [27], [28].
users with a variety of fast response services, The structure of cloud-edge collaboration is
especially in the field of automatic driving generally divided into terminal layer, edge layer
intelligent manufacturing, video monitoring and and cloud computing layer. The following is a brief
other location awareness, rapid feedback is introduction to the composition and functions of
especially important. each layer in the edge computing architecture.
Security: traditional cloud computing requires all
data to be uploaded to the cloud for unified 1) TERMINAL LAYER
processing, which is a centralized processing The terminal layer consists of all types of devices
method. In this process, there will be risks such as connected to the edge network, including mobile
data loss and data leakage, which cannot guarantee terminals and many Internet of Things devices
security and privacy. For example, account (such as sensors, smartphones, smart cars, cameras,
passwords, historical search records and even trade etc.). In the terminal layer, the device is not only a
secrets can all be exposed. Since edge computing is data consumer, but also a data provider. In order to
only responsible for the tasks within its own scope, reduce the terminal service delay, only the
the processing of data is based on the local, there is perception of the various terminal devices is
no need to upload to the cloud, to avoid the risks considered, not the computing power. As a result,
brought by the network transmission process, so hundreds of millions of devices in the terminal
the security of data can be guaranteed. When data layer collect all kinds of raw data and upload it to
is attacked, it only affects local data, not all data. the upper layer, where it is stored and calculated.
Low cost, low energy consumption, low
bandwidth cost: in edge computing, since the data 2) BOUNDARY LAYER
to be processed does not need to be uploaded to the The edge layer is the core of the three-tier
cloud computing center, it does not need to use too architecture. It is located at the edge of the network
much network bandwidth, so the load of network and consists of edge nodes widely distributed
bandwidth is reduced, and the energy consumption between terminal devices and clouds. It usually
of intelligent devices at the edge of the network is includes base stations, access points, routers,
greatly reduced. Edge computing is ‘‘small-scale,’’ switches, gateways, etc. The edge layer supports
and in production,companies canreduce thecost the access of terminal devices downward, and
ofprocessingdata inlocal equipment. Therefore, stores and computes the data uploaded by terminal
edge computing reduces the amount of devices. Connect with the cloud and upload the

20181 VOLUME 8, 2020

K. Cao et al. : Overview on Edge Computing Research

processed data to the cloud [7]. Since the edge As shown in Figure 1, the ECC edge computing
layer is close to the user, the data transmission to reference architecture presents the architecture
the edge layer is more suitable for real-time data content from different perspectives in a multi-view
analysis and intelligent processing, which is more manner, and the functionality of each layer is
efficient and secure than cloud computing. shown through the multi-layer functional
perspective.
3) CLOUD LAYER The edge reference framework has an underlying
Among the federated services of cloud-edge service layer that links the entire framework,
computing, cloud computing is still the most including management services, data lifecycle
powerful data processing center. The cloud services, and security services. Management
computing layer consists of a number of high- services provide unified management, monitoring
performance servers and storage devices, with the operation of the architecture, and providing
powerful computing and storage capabilities, and information to the management platform. The data
can play a good role in areas requiring large lifecycle service provides integrated management
amounts of data analysis such as for the preprocessing, analysis, distribution, and
regularmaintenanceandbusinessdecisionsupport.Th execution of machine data, as well as visualization
ecloud computing center can permanently store the and storage. The security service can define the
reported data of the edge computing layer, and it business logic of the whole life cycle of data
can also complete the analysis tasks that the edge through the business orchestration layer, flexibly
computing layer cannot handle and the processing deploy and optimize the data service, and meet the
tasks that integrate the global information. In real-time requirements of the business. Security
addition, the cloud module can also dynamically services cover all levels of edge computing
adjust the deployment strategy and algorithm of the architecture, adapt to the specific architecture of
edge computing layer according to the control edge computing, and make use of the unified
policy. security management and perception system to
ensure the safe and reliable operation of the entire
B. EDGE COMPUTING REFERENCE FRAME 3.0 architecture.
ECC, jointly initiated by Huawei, Shenyang Fromtheperspectiveofverticalstructure,themodel-
institute of automation, Chinese academy of driven unified service framework is located at the
sciences, China academy of information and top of the scale to realize the development and
communications, and other well-known enterprises, deployment of services. According to the general
put forward the edge computing reference frame framework of edge calculation, it is divided into
3.0 in the edge computing white paper 3.0 released cloud, edge layer and field layer. The edge layer
in December 2018 [29]. The frame of reference is consists of two main parts: the edge node and the
based on model-driven engineering method. In edge manager.
order to model the knowledge of the physical and
digital world, we need to achieve the following
four goals:
1) Establish a real-time and systematic
cognitive model of the physical world and achieve
the cooperation between the physical world and the
digital world;
2) Establish reusable knowledge model
system in each vertical industry based on modeling
method, and complete cross-industry ecological
cooperation;
3) System to system, service to service and
other modelbased interface for interaction, to
achieve decoupling of software interface and
development language, reduce system
heterogeneity;
4) Can effectively support the life cycle of
development service, deployment operation, data
processing and security.

VOLUME 8, 2020 20182

 K. Cao et al.: Overview on Edge Computing Research

FIGURE 1. Edge computing reference architecture 3.0.

Edge node is the entity hardware, which can carry Figure 2 shows the EdgeX Foundry architecture.
the business of edge computing. The edge manager As can be seen from the figure, the ‘‘Southbound’’
mainly uses software to uniformly manage edge contains all IoT application devices that can
nodes. According to hardware characteristics and directly communicate with the edge network. The
service types, edge computing nodes are divided ‘‘Northbound’’ contains the cloud computing
into edge gateways that process and convert center and the communication network with the
network protocols, edge controllers that control
real-time closed-loop services, edge clouds that
process large-scale data, and edges that collect and
process low-cost information Sensors, etc., can
abstract the devices in the edge computing layer
into computing, networking and storage. Next,
implement generic capability calls using
Application Programming Interfaces (API). The
control, analysis and optimization domain function
module is used to realize the information
transmission of upper and lower layers and the
planning of local edge resources. Edge computing
reference architecture 3.0 provides four service
development frameworks from the terminal to the FIGURE 2. EdgeX foundry architecture.
cloud, including lightweight computing systems,
real-time computing systems, intelligent distributed
systems, and intelligent gateway systems. cloud computing center. The ‘‘Southbound’’ is the
source of the data, while the ‘‘Northbound’’ is used
C. EdgeX FOUNDRY to collect the data from the south side and store,
EdgeX Foundry is a neutral open source project integrate and analyze the data. The EdgeX Foundry
hosted by the Linux foundation and is a universal is located between the south and Northbounds and
open framework for computing on the edge of the consists of a collection of microservices that are
Internet of Things. Hosted on a reference software divided into four service layers and two underlying
platform that is completely independent of enhanced system services.
hardware and operating systems, the framework From a vertical perspective, the software
enables a plug-and-play component ecosystem to development kit (SDK) provided by the device
unify the computing open platform at the edge of service layer is used to establish the
the Internet of Things and accelerate deployment of communication links with the ‘‘Southbound’’. The
solutions. device service layer converts the data from the
20183 VOLUME 8, 2020
K. Cao et al. : Overview on Edge Computing Research

device and IV. HOT RESEARCH CONTENT OF EDGE COMPUTING

sendsittothecoreservicelayer.Itcanalsoreceivecomm The emergence of edge computing has promoted
ands from other microservices and pass them to the rapid development of the Internet of Things and
devices. EdgeX Foundry includes multiple access has made a significant contribution to the
methods such as Message Queue Telemetry realization of an intelligent society. Therefore, edge
Transmission Protocol (MQTT), Virtual Device computing has become a hot issue for scientists at
(VIRTUAL), and Bluetooth Low Energy (BLE). home and abroad. This section mainly reviews the
As the core service layer of the center, it is the key key technologies and data security and privacy
to realize edge capabilities. The core service layer protection.
consists of four micro-service components: core
data, commands, metadata, registry, and A. KEY TECHNOLOGIES
configuration. The core data service provides The key technologies of edge computing mainly
storage and management services for device data. include different levels of computing offloading,
The command service is responsible for defining mobility management, traffic offloading
the operation commands in the device file into a technology, caching acceleration, network control
general API, caching and managing commands, and so on.
and can pass the requirements of the cloud
computing center to the device side. Metadata 1) COMPUTING OFFLOADING
provides pairing for devices and services. Computing offloading refers to resource-
Registration and configuration services provide constrained device that partially or fully migrates
configuration information for other micro-services. resource-intensive computing from mobile devices
The support service layer is used to provide edge to resource-rich nearby infrastructure to address
analysis and intelligent services, and to provide the mobile device deficiencies in resource storage,
framework itself with rules engine, scheduling, computing performance, and energy efficiency
alerting and logging services. The application and [33]. The computing offloading technology not
export service layer can be connected with the only reduces the pressure on the core network, but
cloud computing center to transfer data to the cloud also reduces the delay caused by transmission.
computing center to ensure the independent Mobile edge computing (MEC) can run new
operation of EdgeX Foundry. In the export service complex applications on user equipment (UE), and
layer, the client registration service records the computing offloading is the key technology for
relevant information of the back-end registered edge computing. There have been many related
system, and the distribution service exports the research achievements, mainly including two main
corresponding data to the specified client. issues: offloading decision and resource allocation.
Similar to the Edge Computing Reference Among them, the offloading decision is about how
Architecture 3.0, the EdgeX Foundry framework to offload computing tasks, how much to offload
also has a basic service layer that runs through the and what to offload for mobile devices. Resource
whole frameworkmanagement services and allocation is to study where to offload resources.
security services. The management service
provides functions such as installing, upgrading, a: OFFLOADING DECISION
starting, stopping, and monitoring EdgeX Foundry Offloading decision refers to the problem that the
operations. The components in the security service UE decides how to offload the computing task,
are used to protect the data from the device and the how much to offload, and what to offload. In the
operation of the device. offloading system, the UE generally consists of a
The EdgeX Foundry framework was developed code parser, a system parser, and a decision engine.
to simplify and standardize edge computing for the Its execution of the offloading decision is divided
Industrial Internet of Things. It provides an into three steps: (1) The code parser determines
operable open-source platform, in which all micro- what can be offloaded, and the specific offloading
services can run on various operating systems in content depends on the application type and code
the form of containers, and support dynamic data partition; (2) The system parser is responsible
addition or reduction of functions, with strong for monitoring various parameters, such as the
scalability. At present, the application fields of available bandwidth, the size of the data to be
EdgeX Foundry have involved various industries offloaded, or the energy consumed to execute a
such as retail, manufacturing [30], energy [31], local application; (3) Finally, the decision engine
urban parks, transportation [32] and other determines whether to offload or not.
industries.
VOLUME 8, 2020 20184
 K. Cao et al.: Overview on Edge Computing Research

The UE offloading decision results are divided resources and the diversity of networks require
into three cases: local execution, full offloading, adaptive device computing capabilities and changes
and partial offloading. The specific decision results in network bandwidth during the migration process.
are determined by the UE energy consumption and Reference [35] further optimized the virtual
delay in completing computing tasks. According to machine migration strategy by predicting the user’s
the optimization goals of the offloading decision, movement, and proposed a mobilitybased service
the computing offloading can be divided into three migration prediction scheme (MSMP), which
types: reducing latency as the goal, reducing adopted a compromise between cost and service
energy consumption as the goal, and balancing quality.
energy consumption and latency as the goal.
3) OTHER KEY TECHNOLOGIES
b: RESOURCE ALLOCATION In addition to the above two key technologies, the
After completing the offloading decision, we must key technologies of edge computing also include
consider the issue of reasonable resource traffic offloading technology, cache acceleration,
allocation, that is, where to offload. If the and network control.
computing task of the UE is indivisible or can be
divided but the divided parts are related, in this a: TRAFFIC OFFLOADING TECHNOLOGY
case, the offloading task needs to be offloaded to In order to realize the localization, short-distance
the same MEC server; and for the computing tasks deployment, and low-latency, high-bandwidth
that can be divided but not related to the divided transmission capabilities of business applications in
part, it can be offloaded to multiple MEC servers. wireless networks, wireless networks have the
At present, resource allocation nodes are mainly capability of traffic offloading. Offloading (traffic
divided into single-node allocation and multi-node offloading) of edge network traffic is very
allocation. important in mobile edge computing. Traffic
offloading is used to offload traffic that meets
2) MOBILITY MANAGEMENT specific offloading rules to mobile edge networks
Edge computing relies on the geographical (that is, a local specific network, which can be an
distribution of resources to support the mobility of intranet or the Internet) to save backhaul
applications. An edge computing node only serves bandwidth, reduce latency, and facilitate the
users around it. The cloud computing mode expansion of other MEC services [36]. Reference
supports the application mobility by fixing the [37] proposed a method of energyefficient traffic
location of the server and transmitting the data to offloading for mobile users in a two-layer
the server through the network, so the mobile heterogeneous wireless network. Experimental
management of the application in the edge results show that the method can save up to 34
computing is a new mode. The main issues percents of energy under typical network settings.
involved are resource discovery and resource
switching. b: CACHING ACCELERATION
Resource discovery, that is, users need to quickly Mobile edge caching technologies include base
discover the resources available around and choose station caching, mobile content distribution
the most suitable resources during the movement. networks, and transparent caching. Caching
The resource discovery of edge computing needs to acceleration technology can improve the efficiency
adapt to the heterogeneous resource environment, of content distribution and improve the user
and also needs to ensure the speed of resource experience. After the content is cached to the edge
discovery, so that applications can provide services of the mobile network, users can obtain the content
to users without interruption; resource switching, nearby, thereby avoiding repeated transmission of
that is, when users move, the computing resources content, and alleviating the pressure on the
used by mobile applications may be switched backhaul network and core network. At the same
among multiple devices [34]. Resource switching time, the edge caching can reduce the network
will migrate the service program’s operation site to delay requested by the user, thereby improving the
ensure service continuity. user’s network experience. In addition, the edge
In MEC, one of the key issues to be considered caching can also open up the mobile network
is how to ensure the continuity of users’ access to resource environment and provide more abundant
services during the movement. Some applications services for tenants and users [38]. Reference [39]
expect to continue serving users after their location proposed a cognitive agent (CA) to help users
changes. The heterogeneity of edge computing cache and perform tasks on the MEC in advance,
20185 VOLUME 8, 2020
K. Cao et al. : Overview on Edge Computing Research

and to coordinate communication and cache to ease scope of data authorization. In addition, due to the
the pressure on the MEC. outsourcing nature of data, its ownership and
control are separated from each other, so an
effective audit verification scheme can ensure the
c: NETWORK CONTROL
integrity of the data.
An edge network is a given non-technically
(3) Security challenges between large-scale
described network (the edge of a public
interconnected services for edge computing and
telecommunication network). The edge network
resource-constrained terminals. Due to the multi-
includes part or all of the aggregation layer
source data fusion characteristics of edge
network and the access layer network, and is the
computing, the superposition of mobile and
last segment of the network for accessing users. In
Internet networks, and the resource limitations of
terms of value, edge networks are commercial
storage, computing, and battery capacity of edge
networks between existing core networks and large
terminals, traditional and more complex encryption
users. In terms of network control, reference [40]
algorithms, access control
proposed an effective workload slicing scheme, in
measures,identityauthenticationprotocolsandprivac
which users use software-defined networks to
yprotection methods cannot be applied in edge
processdataintensive applications in multi-edge
computing.
cloud environments.
(4) Diversified services for the Internet of
Things and the new requirements of edge
B. DATA SECURITY AND PRIVACY PROTECTION The computing mode for efficient privacy protection. In
security of edge computing is one of the hot addition to the need to design effective data,
research issues. Network edge data involves location and identity privacy protection schemes,
personal privacy. Although the concept of data how to combine traditional privacy protection
processing nearby also provides better structured schemes with edge data processing characteristics
support for data security and privacy protection, the in edge computing environments to enable user
distributed architecture of edge computing privacy protection in a diverse service environment
increases the dimension of attack vectors. The is the future research trend.
smarter the edge computing client, the more At present, the research on edge computing
vulnerable they are to malware infections and security and privacy protection is still in its
security breaches. Existing data security protection infancy, and there are relatively few existing
methods are not fully applicable to edge computing research results. Among them, a really feasible
architectures. Moreover, the highly dynamic research idea is to port existing security
environment at the edge of the network also makes technologies in other related fields to the edge
the network more vulnerable and difficult to computing environment. Scholars at home and
protect. Data security and privacy protection in abroad have carried out in-depth research on
edge computing faces four new challenges: mobile cloud computing and its security. Roman et
(1) New requirements for lightweight data al. [41] conducted security analysis on several
encryption and fine- common mobile edge paradigms, elaborated a
graineddatasharingbasedonmultipleauthorizedparti general cooperative security protection system, and
es in edge computing. Because edge computing is a gave research opinions. These works provide
computing mode that integrates multiple trust theoretical reference for the security research of
domains with authorized entities as trust centers, edge computing. This paper divides the research
traditional data encryption and sharing strategies system of data security and privacy protection in
are no longer applicable. Therefore, it is edge computing into four
particularly important to design a data encryption parts:datasecurity,identityauthentication,privacypro
method for multiple authorization centers. At the tection and access control.
same time, the complexity of the algorithm should
be considered. 1) DATA SECURITY
(2) Multi-source heterogeneous data Data security is the foundation of creating a secure
propagation control and security management edge computing environment, whose fundamental
issues in a distributed computing environment. purpose is to ensure the confidentiality and
Users or data owners want to be able to use integrity of data. It is mainly aimed at the
effective information dissemination control and characteristics of separation of ownership and
access control mechanisms to achieve data control of outsourced data and randomization of
distribution, search, access, and control of the storage, and is used to solve problems such as data
VOLUME 8, 2020 20186
 K. Cao et al.: Overview on Edge Computing Research

loss, data leakage, and illegal data operations. At solution should have the ability to perform batch
the same time, on this basis, users are allowed to audits.
perform secure data operations. So far, most of the 3) Privacy protection. Since neither the data
research results of scholars at home and abroad storage server nor the data owner is suitable for
have focused on cloud computing [42], mobile performing integrity audit schemes, they often
cloud computing [43] and fog computing [44]. need to be built with a third-party auditing platform
Therefore, a main research idea of data security in (TPA). In this case, when the TPA is semitrusted
edge computing is to migrate data security or untrusted, security threats such as data leakage
solutions in other computing paradigms to the edge and tampering are very likely, and data privacy
computing paradigm, and to parallelize the cannot be guaranteed. Therefore, protecting user
distributed computing architecture in edge data privacy during the integrity audit process is
computing, limited terminal resources, edge big essential.
data processing, highly dynamic environment and 4)Lowcomplexity.Fordatastorageservers(edgedat
other characteristics are organically combined to acenters) and data owners (edge devices) have
finally achieve a lightweight and distributed data limitations in terms of computing power, storage
security protection system. capacity, network bandwidth, etc., in addition to
ensuring data integrity when designing an integrity
a: DATA CONFIDENTIALITY AND SECURE DATA SHARING audit scheme, the scheme is complex The issue of
Existing data confidentiality and secure data degree is also an important factor.
sharing solutions are usually implemented by
encryption technology. c: SEARCHABLE ENCRYPTION
The conventional process is that the data owner In the traditional cloud computing paradigm, in
encrypts and uploads the outsourced data in order to achieve data security and reduce terminal
advance, and the data user decrypts the data when resource consumption, users often use some
necessary. Traditional encryption algorithms encryption method to outsource file encryption to a
include symmetric encryption algorithms (such as third-party cloud server. However, when users need
DES, 3DES, ADES, etc.) and asymmetric to find related files that contain a certain keyword,
encryption algorithms (such as RSA, Diffe- they will encounter the difficulty of how to perform
Hellman, ECC, etc.), but the operability of data a search operation on the cipher text of the cloud
encrypted by traditional encryption algorithms is server. Therefore, searchable encryption (SE) came
low, which causes great obstacles to the subsequent into being. SE can guarantee the privacy and
data processing. At present, the commonly used availability of data, and support the query and
data encryption algorithms include attribute based retrieval of ciphertext data. Similarly, in the edge
encryption (ABE) [45], proxy re encryption (PRE) computing paradigm, user’s file data will be
[46], and all homomorphic encryption (FHE) [47], encrypted and outsourced to the edge computing
etc. center or cloud server. Searchable encryption is
also an important method to protect user privacy in
b: INTEGRITY AUDIT edge computing.
After the user’s data is stored in the edge or cloud
data center, an important issue is how to determine d: SUMMARY
the integrity and availability of outsourced storage 1) In terms of data confidentiality and secure
data. The current research on data integrity audits data sharing, combining with application
has focused on the following four functional encryption theory such as attribute encryption,
requirements [48]: proxy re-encryption, and homomorphic encryption,
1) Dynamic audit. User data in the data how to design a low-latency, distributed secure
storage server is often dynamically updated. The storage system that supports dynamic operations
common dynamic data operations include and correctly handle network edge devices The
modification, copying, inserting, and deleting. synergy between cloud centers is an important
Therefore, the data integrity audit scheme cannot research idea.
be limited to static data, but should have dynamic 2) In the field of data integrity auditing, one
audit capabilities. of the main research goals is to improve auditing
2) Batch audit. When a large number of efficiency and reduce verification overhead while
users issue audit requests at the same time or data realizing various auditing functions. Secondly,
is stored in multiple data centers in blocks, in order designing an integrity audit scheme that supports
to improve audit efficiency, the integrity audit multi-source heterogeneous data and dynamic data
20187 VOLUME 8, 2020
K. Cao et al. : Overview on Edge Computing Research

updates is expected to become the focus of future on (SSO) authentication mechanism is expected to
research. be applied to identity authentication between
3) In terms of searchable encryption, first, multiple trust domains [49]. Reference [50]
how to construct a keyword-based search scheme designed an attribute-based authentication and
under the distributed storage service model and authorization framework for structured P2P
further expand it into the edge computing networks. The framework uses attribute certificates
environment is a feasible research idea; second, and distributed certificate revocation systems to
how to implement it in a secure multi-party sharing replace the traditional P2P network’s public key
mode Fine-grained search permission control certificates and access control list authentication
makes it suitable for multi-user search mechanisms. A server or third-party trusted
environments with different trust domains, while authority that achieves flexible, efficient, and
ensuring search speed and accuracy. Finally, for privacyprotected rights assignments without
the distributed ciphertext data storage model in requiring any external intervention. Literature [51]
edge computing, how to efficiently construct a proposed a cross-domain dynamic anonymous
secure index suitable for resource-constrained group key management authentication system (CD-
network edge devices and design a distributed AGKMS). This system achieves cross-domain
searchable encryption algorithm is an urgent group density by establishing a tree-level hierarchy
problem. with the key generation center (KGC) as the top
layer. Key negotiation. At the same time, in terms
2) IDENTITY AUTHENTICATION of group key management, the scheme provides a
To use the computing services provided by edge time-controlled key revocation mechanism, and the
computing, IoT users must first perform identity user’s key is revoked when the validity period
authentication. Because edge computing is a expires. In addition, CD-AGKMS does not require
distributed interactive computing environment the calculation of bilinear pairs, which improves
where multiple trust domains coexist, it is the feasibility and efficiency of the system.
necessary not only to assign an identity to each
entity, but also to consider mutual authentication c: SWITCHING AUTHENTICATION
between different trust domains. The main research Due to the high mobility of terminal equipment in
content of identity authentication includes identity edge computing, the geographic location of mobile
authentication within a single domain, cross- users often changes, making the traditional
domain authentication, and handover centralized authentication protocol no longer
authentication. applicable to such situations. Handover
authentication is a kind of authentication handover
a: IDENTITY AUTHENTICATION IN A SINGLE DOMAIN
technology to solve the problem of high mobility
The identity authentication in a single trust domain
user identity authentication. Therefore, the research
is mainly used to solve the identity allocation
on handover authentication technology can provide
problem of each entity. Each entity must first pass
strong guarantee for realtime and accurate
the security authentication of the authorization
authentication of edge devices in edge computing.
center to obtain storage and computing services.
At the same time, the privacy of user identity in the
With the deepening of research, designing identity
process of authentication handover is also a
authentication protocols with privacy protection
research focus.
features is the focus of current research.

b: CROSS-DOMAIN AUTHENTICATION d: SUMMARY

At present, the research on the authentication At present, domestic and foreign researchers have
mechanism applicable to entities in different trust mostly improved and optimized the authentication
domains is still in its infancy, and a relatively protocols based on the existing security protocols,
complete research context and theoretical methods including the flexibility, high efficiency, energy
have not yet been formed. In the research of saving and privacy protection of the protocols. In
identity authentication in cloud computing, identity edge computing, the research of identity
management between multiple cloud service authentication protocol should draw on the
providers can be regarded as a form of cross- advantages of existing schemes, and at the same
domain authentication, which makes some time combine the characteristics of distribution and
authentication standards applicable to multi-cloud mobility in edge computing to strengthen the
(such as SAML, OpenID, etc.) And the single sign- research of unified authentication, cross-domain
VOLUME 8, 2020 20188
 K. Cao et al.: Overview on Edge Computing Research

authentication and handover authentication server and service provider are separated to resist
technologies to ensure the data and privacy security illegal access IDM and traffic interception attacks.
of users in different trust domains and At the same time, an additional authentication layer
heterogeneous network environments. is added to prevent mobile devices from
3) PRIVACY PROTECTION compromising.
Not all authorized entities in edge computing are In addition, in order to improve the performance
trusted, but the user’s identity information, location and security of the scheme, the credential
information and private data are stored in these information of mobile devices will be updated in
semi-trusted entities, which easily leads to privacy real time according to the mobile cloud packet
problems. At present, the research on privacy switching mechanism to prevent credential theft
protection is mainly concentrated in the attacks.
environment of mobile cloud and fog computing.
Therefore, privacy protection is a research system d: SUMMARY
that has attracted much attention in edge computing The privacy problems of users in edge computing
based on open interconnection. Its main contents can be summarized into the following three
include data privacy protection, location privacy contradictions: 1) Contradiction between
protection and identity privacy protection. Outsourcing Data and Data Privacy; 2) The
contradiction between location based service and
a: DATA PRIVACY PROTECTION location privacy; 3) The contradiction between data
Because of the user’s privacy data will be stored sharing and identity privacy protection. Scholars at
and processed by entities that are not under the home and abroad have carried out in-depth research
user’s control. Therefore, it is the current research to solve these three contradictions, but the
focus to allow users to perform various operations proposed scheme still has many defects, and some
(such as auditing, searching and updating) on data possible research directions are as following.
while ensuring the privacy of users is not leaked. 1) Support users to perform various
operations (such as auditing, searching and
b: LOCATION PRIVACY PROTECTION updating) on data while ensuring the privacy of
With the popularity of location based service, the users is not leaked.
issue of location privacy has also become a In addition, the privacy issue in the cooperative
research focus. At present, the research focus in interoperability among users deserves extensive
this field mainly focuses on the use of K- attention.
anonymity technology to achieve privacy 2) In view of the shortcomings of TTP-based
protection in location services. However, the privacy protection scheme in computing energy
location privacy protection scheme based on K- consumption, it is particularly important to design
Anonymity consumes a large amount of network a lightweight and efficient privacy protection
bandwidth and computational overhead in practical scheme.
applications, and is not suitable for edge devices 3) A large amount of real-time dynamic data
with limited resources. will be generated by edge devices in the actual
network, which provides the possibility of data
c: IDENTITY PRIVACY PROTECTION correlation, integration analysis and privacy mining
At present, the protection of user identity privacy for attackers. Therefore, it is an important research
in the edge computing paradigm has not attracted content to construct dynamic and fine-grained data
widespread attention, only some exploratory security and privacy protection schemes from the
research results in the mobile cloud environment. perspective of user’s identity, behavior, interest
Khalil et al. [52] pointed out that the current third- and location.
party identity management system (IDM) is
4) ACCESS CONTROL
vulnerable to three kinds of attacks: IDM server
compromise, mobile device compromise and Access control is a key technology and method to
network traffic interception. Aiming at these ensure system security and protect user privacy.
attacks, this paper proposes a comprehensive Currently, popular access control schemes include
thirdparty identity management system (CIDM), attribute-based and rolebased access control.
which manages mobile users’ digital identities on Among them, attribute-based access control can be
behalf of service providers by introducing IDM well applied to distributed architecture and realize
servers. Firstly, the authorization certificate, IDM fine-grained data sharing.

20189 VOLUME 8, 2020

K. Cao et al. : Overview on Edge Computing Research

a: ATTRIBUTE-BASED ACCESS CONTROL Internet traffic. According to the forecast [53], the
Because edge computing is a data-oriented total global IP traffic will triple between 2016 and
computing mode, access control of edge computing 2021, from an annual average of 1.2ZB(96EB/
is usually implemented by cryptographic month) in 2016 to 3.3ZB(278EB/ month) in 2021.
technology. Traditional cryptographic technology Figure 3 shows the statistics and forecast of the
is not suitable for distributed parallel computing annual monthly average of global internet traffic.
environment, while Attribute Encryption (ABE) The proportionof video in Internet traffic
can be well applied to distributed architecture to willincrease from 73 percent in 2016 to 82 percent,
realize fine-grained data sharing and access control. and it can be predicted that this proportion will
increase year by year.
b: ROLE-BASED ACCESS CONTROL
Role-based access control provides flexible control
and management through a dual privilege mapping
mechanism, that is, the privilege mapping from
users to roles and roles to data objects.

c: SUMMARY
To sum up, access control technology is the key
technology and important method to ensure system
security and protect user privacy. In principle, the
access control system in edge computing should be
applicable to multi-entity access control between
different trust domains, and various factors such as FIGURE 3. Annual monthly average statistics/forecast of global internet
geographic location and resource ownership should traffic.
also be considered. Therefore, the design of a fine-
grained, dynamic, lightweight and multi-domain The increasing video data traffic year by year
access control mechanism is the focus of the next will occupy more Internet bandwidth resources. In
research, and an efficient access control method front of the limited bandwidth resources, it is
based on attributes and roles should be a more undoubtedly an effective scheme to reasonably use
suitable technical means for edge computing the edge computing platform to cache local video.
environment. MEC, a platform based on edge computing with
video analysis and caching functions, is deployed
V. EDGE COMPUTING APPLICATION SCENARIOS in areas with high traffic such as university towns,
The birth of each emerging technology is followed residential areas, commercial streets and others and
by its corresponding application in different frequent requests for video playback. The edge
scenarios. The important criterion to test the computing intelligent analysis function (based on
feasibility of the new technology is whether it is search heat) [54] is used to cache popular TV
efficient to solve the existing problems in the actual plays, movies and other video resources with high
environment. The various challenges and download frequency on the nearby MEC server.
opportunities that edge computing will face in the When a user sends out a video playing request, the
application process are presented. With the video resource can achieve the effect of loading
improvement of edge computation in theory, more from the local, thus not only saving bandwidth, but
and more applications based on edge computation also greatly reducing the waiting time of the user.
are called reality. This section will deeply In addition, content in MEC platform is optimized
understand edge calculation through the application based on RAN-side perception, so that content can
of edge calculation in five scenes: edge calculation be dynamically optimized according to network
video cache scene, 5G communication scene, edge real-time information (network load, link quality,
calculation network video live broadcast scene, data throughput rate, etc.), and the effect of
predictive maintenance scene and security improving Quality of Experience (QoE) and
monitoring scene. network efficiency is achieved.
A. EDGE COMPUTING VIDEO CACHE
B. EDGE CALCULATION AND 5G
With the rapid development of the emerging The upcoming commercialization of the fifth
Internet industry, the booming application generation mobile communication network (5G
programs have led to the continuous growth of network) provides new opportunities for the

VOLUME 8, 2020 20190

 K. Cao et al.: Overview on Edge Computing Research

development of edge computing. 5G has the vehicles for code distribution based on machine
advantages of small delay, large bandwidth and learning [57], etc.
large capacity, which solves many problems
encountered in the traditional communication field, C. EDGE COMPUTING NETWORK VIDEO LIVE
but also leads to the rapid growth of data volume. BROADCAST
At this time, it is urgent to provide a reliable, useful The network video live broadcast system is a
and executable business model. The characteristics multimedia network platform, which aims to
of 5G, such as fast processing and low latency, can transmit live audio and video live events such as
provide a new way for rapid response, and can ongoing competitions, conferences, performances
jointly optimize the end, edge and cloud. This and teaching to remote audiences in real time
capability of edge computing can intelligently through the network. Although the server port of
allocate resources among Internet of Things the traditional live video broadcasting system
devices, edge devices and cloud devices from generally adopts 100 megabytes or gigabytes of
aspects of user experience, power consumption, network, due to the large audio andvideo
computing load, performance, cost, etc., providing files,thedelay probleminthe wholeprocesscannot be
a new approach for joint optimization. Therefore, ignored.
the development of edge computing technology is To solve the above problems, Shanghai
closely related to 5G: on the one hand, edge Mercedes-Benz Cultural Center has introduced
computing can support 5G, and the important multi-access edge computing technology. Edge
component of 5G is edge computing. On the other nodes are deployed by adopting the technical
hand, because 5G is expressed in the form of scheme of China Unicom edge video scheduling
software, edge calculation can be flexibly applied network. Video shot in the venue is stored in a
[4]. dedicated edge cloud, and audience in the venue
In the European market, the edge computing can access the video information stored in the edge
industry has formed an industrial alliance, with cloud through mobile devices, thus avoiding the
large technology enterprises represented by time delay caused by connecting to the central
Vodafone, Deutsche Telekom, Siemens and other cloud. The EVO solution can control the delay of
companies already joining in. The European live network video broadcasting to millisecond
Telecommunications Standards Institute (ETSI) has level, which is more than 60 times lower than the
initiated the formulation of standardized Mobile average delay of ordinary live network video
Edge Computing (MEC). Operators can open their broadcasting, and can support hundreds of millions
wireless network edges to authorized third parties of off-site Internet viewers to watch high-definition
so that they can flexibly and quickly deploy live broadcast.
innovative applications and services for mobile
users, enterprises and vertical network segments D. PREDICTIVE MAINTENANCE
[8], [9]. Mobile edge computing is the result of the A reliable and efficient maintenance plan is of vital
natural development of mobile base station importance to manufacturers, because passive
iteration and the integration of IT and shutdown caused by equipment failure may have a
telecommunication networks. It will provide new great impact on production efficiency and safety.
vertical service for consumers and enterprise Predictive Maintenance [58] is one of the key
customers, including video analysis, location innovations proposed by Industry 4.0. It forecasts
service, Internet of Things, augmented reality, and optimizes the maintenance of equipment
optimization of local content distribution and data operation based on continuous measurement and
caching, etc. In February 2018, ETSI released two analysis of the system.
white papers, namely ‘‘Cloud RAN and Mobile At present, most manufacturers use preventive
Edge Computing: Perfect Pairing’’ and maintenance to improve the stability of production
‘‘Deployment of Mobile Edge Computing in 4G lines. Preventive maintenance is usually carried out
and Evolution to 5G’’, in order to keep mobile on a time basis. Through regular maintenance, the
edge computing synchronized with 5G. probability of equipment failure or shutdown
In addition, people are also applying 5G to the within a period of time is reduced. Preventive
Internet of Things, such as the Internet of Things maintenance can reduce the total downtime of the
network management system [55], the intelligent production line and the number of failures caused
resource allocation management of the vehicle by equipment deterioration, and has the advantages
network [56] and the selection of reliable mobile of convenient implementation and strong
operability. However, as the maintenance time is
20191 VOLUME 8, 2020
K. Cao et al. : Overview on Edge Computing Research

determined by experience, there may be nodes and cloud computing centers, and can meet
insufficient or excessive maintenance. With the the requirements of predictive maintenance for
maturity of Internet of Things, big data and other real-time and privacy protection, while ensuring the
technologies, predictive maintenance makes accuracy of fault prediction.
maintenance more intelligent by analyzing real-
time monitoring data of equipment, predicting E. SECURITY MONITORING
possible failures of Vision is one of the important ways for human
equipment,andproposingthecausesandsolutionsoffai beings to know the world and obtain information.
lures. The key to predictive maintenance is to The ‘‘Skynet’’ monitoring system deployed by our
troubleshoot hidden troubles in advance and solve public security organs maintains a stable and safe
them, which can reduce the total maintenance cost, social order through a large number of cameras
failure rate and total downtime and improve the deployed in public areas. Many families also take
reliability of the equipment. However, due to the the initiative to use household cameras and pet
large number and variety of terminals, the practical monitors to protect the safety of their houses and
application of predictive maintenance faces the family members. At first, people can only process
problems of handling the connection and image information manually, but this method often
management of mass terminals, ensuring the real- has a long time delay and fluctuation accuracy.
time analysis and protecting the privacy of With the continuous development of artificial
industrial data. intelligence technology, image data computer
According to the American Journal of Efficient already has very strong learning and processing
Plant, the emergence of edge computation is of capabilities. However, the traditional cloud
great significance to the implementation of computing model is difficult to support the
predictive maintenance schemes. The strong application of image processing in some scenes.
perception ability of edge nodes and their close This paper takes railway track foreign body
proximity to equipment can meet the real-time and detection as an example to elaborate in more detail.
privacy protection requirements of predictive First of all, since the railway tracks are laid
maintenance. Matt Boujonnier, an analysis outdoors and include tunnels, mountains and other
application architecture engineer at Schneider areas where network quality cannot be guaranteed,
Electric, pointed out that machine learning the use of cloud computing models is likely to
algorithms can usually only be run in cloud result in image loss or serious quality loss.
computing centers, but in Internet of Things Secondly, rail foreign matter detection requires
applications, people hope that algorithms can be very high real-time performance. Danger must be
run wherever necessary. The Realift Rod Pump found in time and alarm must be made. Cloud
controller jointly developed by Schneider Electric computing processing link is long, and there is no
and Microsoft Azure has realized real-time analysis guarantee of network bandwidth, so the real-time
and prediction of the operating state of the performance of detection and alarm may not be
equipment at the edge of the network and has been guaranteed. At the same time, rail monitoring
applied in the oilfield industry on a pilot basis. images have certain confidentiality, once uploaded
Huawei also pointed out that Internet of Things to the cloud, there is a risk of theft and tampering
based on edge computing (e.g., Edge Computing- for criminals. Finally, the length of China’s railway
IoT) can effectively construct predictive is 121,000 kilometers. If all the monitoring image
maintenance solutions, and has launched services data are uploaded to the cloud server for analysis, it
for designing and deploying predictive is bound to require the cloud server to have
maintenance solutions. Huawei uses intelligent extremely strong processing and computing
gateways to provide intelligent services, real-time capabilities. Obviously, image processing based on
monitoring and analysis of key indicators of edge computation can provide better services in
maintenance objects, prediction of possible failures scenes with high real-time requirements, network
of maintenance objects, and information reporting. quality cannot be guaranteed and privacy is
The cloud computing center performs a involved.
comprehensive state evaluation based on the Hikvision is a provider of Internet of Things
comprehensive information of multiple objects, and solutions with video as the core, and has been
can continuously iterate and optimize the deeply engaged in security monitoring for many
prediction algorithm run by edge nodes to realize years. In October 2017, Hikvision first publicly
dynamic deployment. This maintenance scheme released the Cloud Edge Collaboration Architecture
comprehensively utilizes the advantages of edge oftheAICloud,whichconsistsofaCloudComputingCe
VOLUME 8, 2020 20192
 K. Cao et al.: Overview on Edge Computing Research

nter, an Edge Domain and an Edge Node, enabling research issue. In the future, with the continuous
AI reasoning capability to edge, thus realizing fast development of the Internet and human society,
and efficient perception, while the Cloud edge computing will play a more important role
Computing Center focuses on global cognition and and effectively promote the development of various
analysis. ‘‘Deep Eyes’’ binocular behavior analysis industries. It plays an important application role in
camera is a representative edge intelligent product Content Delivery Network(CDN), industrial
of Hikvision. It has built-in high-performance GPU Internet, energy, smart home, smart transportation,
and artificial intelligence algorithm. It can also games and other fields.
analyze and detect 9 behaviors such as crossing the
warning line, wandering, running and leaving the REFERENCES
post in an offline state. It can be applied to bank [1] D. Evans. The Internet of Things How The Next Evolution of
vaults, guard rooms, hospitals and other scenes. the Internet is Changing Everything. Accessed: Dec. 3, 2016.
[Online]. Available:
Academia is also optimistic about the application https://www.researchgate.net/publication/30612290
of edge computing in [59]. The system architecture [2] V. Turner, J. F. Gantz, and D. Reinsel. (Nov. 26, 2018). The
consists of a monitoring application layer, a fog digital universe of opportunities: Rich Data and the
computing layer and a cloud computing layer. The Increasing Value of the Internet of Things. [Online].
Available: https://www.emc.com/leadership/
fog computing layer includes cameras, smart digitaluniverse/2014iview/index.htm
tablets, smart phones and other devices. Fog nodes [3] Y. Q. Gao, H. Bguan, and Z. W. Qi, ‘‘Service level agreement
can track suspicious targets and calculate the based energy-Effificient resource man agreement in cloud data
centers,’’ Comput. Elect. Eng., vol. 40, no. 5, pp. 1621–1633,
driving speed in real time. The processing results 2014, doi: 10.1016/j. compeleceng.2013.11.001.
will be sent to the cloud computing center. This [4] M. Satyanarayanan, ‘‘The emergence of edge computing,’’
mode can greatly reduce network traffic and Computer, vol. 50, no. 1, pp. 30–39, Jan. 2017.
improve the real-time performance of the system. [5] W. S. Shi, X. Z. Zhang, and Y. F. Wang, ‘‘Edge computing:
State-of-the-art and future directions,’’ J. Comput. Res.
Because the resources of edge intelligence Develop., vol. 56, no. 1, pp. 1–21, 2019.
equipment are usually limited, and artificial [6] W. Shi, H. Sun, J. Cao, Q. Zhang, and W. Liu, ‘‘Edge
intelligence algorithms based on deep learning computing-an emerging computing model for the Internet of
everything era,’’ J. Comput. Res. Develop., vol. 54, no. 5, pp.
often need a lot of computing and storage resources
907–924, May 2017.
to run, so the lightweight algorithm is also one of [7] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, ‘‘Edge computing:
the research hotspots in the application of edge Vision and challenges,’’ IEEE Internet Things J., vol. 3, no. 5,
computing in security monitoring. In 2016 pp. 637–646, Oct. 2016.
[8] Z. M. Zha, F. Liu, and Z. P. Cai, ‘‘Edge computing: Platforms;
DeepScale Company proposed a lightweight target Applications and challenges,’’ J. Comput. Res. Develop., vol.
detection neural networkSqueeze Net [60]. After 55, no. 2, pp. 327–337, 2018.
compression, the network volume is as low as 0.5 [9] X. Hong and Y. Wang, ‘‘Edge computing technology:
MB, but this algorithm cannot meet the real-time Development and countermeasures,’’ Chin. J. Eng. Sci., vol.
20, no. 2, p. 20, 2018.
detection. In 2018, scholars from Binghamton [10] ‘‘The Internet of Things reference model,’’ in Internet of
University used a lightweight real-time detection Things World Forum, Oct. 14. Chicago, IL, USA: C. S. Inc,
and tracking algorithm in reference [61] to realize 2014, pp. 1–12.
[11] X. Sun and N. Ansari, ‘‘Edge IoT: Mobile edge computing for
pedestrian recognition, tracking and abnormal
the Internet of Things,’’ IEEE Commun. Mag., vol. 54, no. 12,
behavior detection in public areas. Kerman pp. 22–29, Dec. 2016.
algorithm [62] was proposed in June 2018. It uses a [12] A. Alrawais, A. Alhothaily, C. Hu, and X. Cheng, ‘‘Fog
hybrid filter based on decision tree to construct a computing for the Internet of Things: Security and privacy
issues,’’ IEEE Internet Comput., vol. 21, no. 2, pp. 34–42,
lightweight convolution neural network for human Mar. 2017.
target tracking. [13] J. Kang, R. Yu, X. Huang, and Y. Zhang, ‘‘Privacy-preserved
pseudonym scheme for fog computing supported Internet of
vehicles,’’ IEEE Trans. Intell. Transp. Syst., vol. 19, no. 8, pp.
VI. CONCLUSION
2627–2637, Aug. 2018.
This paper systematically introduces the edge [14] C. Mouradian, D. Naboulsi, S. Yangui, R. H. Glitho, M. J.
computing model from the aspects of basic Morrow, and P. A. Polakos, ‘‘A comprehensive survey on fog
concepts, architecture, key technologies, security computing: State-of-theArt and research challenges,’’ IEEE
Commun. Surveys Tuts., vol. 20, no. 1, pp. 416–464, 1st
and privacy protection. Edge computing provides Quart., 2018.
data storage and computing at the edge of the [15] Z. M. Xu and Y. F. Tian, ‘‘The history and application of
network, and provides Internet intelligent services cloud computing,’’ Inf. Recording Mater., vol. 19, no. 8, pp.
66–67, 2018.
nearby, providing support for the digital
[16] L. Jiao, R. Friedman, X. M. Fu, S. Secci, Z. Smoreda, and H.
transformation of various industries, and meeting Tschofenig, ‘‘Cloud based computation off loading for mobile
the requirements of different industries for data devices: State of the art, challenges and opportunities,’’
diversification. Edge computing has become a hot
20193 VOLUME 8, 2020
K. Cao et al. : Overview on Edge Computing Research

Future Netw. Mobile Summit, Lisbon, Portugal, Jul. 2013, pp. [36] L. I. N. Kaiqun and C. T. C. F. Branch, ‘‘Pilot study on
1–11. application of MEC local shunting service,’’ Modern Inf.
[17] I. Stojmenovic, ‘‘Fog computing: A cloud to the ground Technol., vol. 1, no. 3, pp. 65–67, 2017.
support for smart things and machine-to-machine networks,’’ [37] F. Lu, J. Hu, L. T. Yang, Z. Tang, P. Li, Z. Shi, and H. Jin,
in Proc. Australas. Telecommun. Netw. Appl. Conf. (ATNAC), ‘‘Energyefficient traffic offloading for mobile users in two-tier
Nov. 2014, pp. 117–122. heterogeneous wireless networks,’’ Future Gener. Comput.
[18] S. Yangui, P. Ravindran, O. Bibani, R. H. Glitho, N. Ben Syst., vol. 105, pp. 855–863, Apr. 2020.
Hadj-Alouane, M. J. Morrow, and P. A. Polakos, ‘‘A platform [38] D. Zhu and X. D. Wang, ‘‘Mobile network edge computing
as-a-service for hybrid cloud/fog environments,’’ in Proc. and caching technology,’’ Railway Comput. Appl., vol. 26, no.
IEEE Int. Symp. Local Metrop. Area Netw. (LANMAN), Jun. 8, pp. 51–54, 2017.
2016, pp. 1–7. [39] R. Wang, M. Li, L. Peng, Y. Hu, M. M. Hassan, and A.
[19] X. Zhu, D. S. Chan, and M. S. Prabhu, ‘‘Improving video Alelaiwi, ‘‘Cognitive multi-agent empowering mobile edge
performance with edge servers in the fog computing computing for resource caching and collaboration,’’ Future
architecture,’’ Intel Technol. J., vol. 19, no. 1, pp. 202–224, Gener. Comput. Syst., vol. 102, pp. 66–74, Jan. 2020.
2015. [40] G. S. Aujla, N. Kumar, A. Y. Zomaya, and R. Ranjan,
[20] J. L. Zhang, Y. C. Zhao, and B. Chen, ‘‘Survey on data ‘‘Optimal decision making for big data processing at edge-
security and privacypreserving for the research of edge cloud environment: An SDN perspective,’’ IEEE Trans. Ind.
computing,’’ J. Commun., vol. 39, no. 3, pp. 1–21, 2018. Informat., vol. 14, no. 2, pp. 778–789, Feb. 2018.
[21] Q. Qi and F. Tao, ‘‘A smart manufacturing service system [41] R. Roman, J. Lopez, and M. Mambo, ‘‘Mobile edge
based on edge computing, fog computing, and cloud computing: A survey and analysis of security threats and
computing,’’ IEEE Access, vol. 7, pp. 86769–86777, 2019. challenges,’’ Future Gener. Comput. Syst., vol. 78, pp. 680–
[22] L. Ruan, Y. Yan, S. Guo, F. Wen, and X. Qiu, ‘‘Priority-based 698, 2018.
residential energy management with collaborative edge and [42] D. Zissis and D. Lekkas, ‘‘Addressing cloud computing
cloud computing,’’ IEEE Trans. Ind. Informat., vol. 16, no. 3, security issues,’’ Future Gener. Comput. Syst., vol. 28, no. 3,
pp. 1848–1857, Mar. 2020. pp. 583–592, Mar. 2012.
[23] T. Wang, G. Zhang, A. Liu, M. Z. A. Bhuiyan, and Q. Jin, ‘‘A [43] H. Liang, D. Huang, L. X. Cai, X. Shen, and D. Peng,
secure IoT service architecture with an efficient balance ‘‘Resource allocation for security services in mobile cloud
dynamics based on cloud and edge computing,’’ IEEE computing,’’ in Proc. IEEE Conf. Comput. Commun.
Internet Things J., vol. 6, no. 3, pp. 4831–4843, Jun. 2019. Workshops (INFOCOM WKSHPS), Apr. 2011, pp. 191–195.
[24] M. S. Hossain and G. Muhammad, ‘‘Emotion recognition [44] I. Stojmenovic and S. Wen, ‘‘The fog computing paradigm:
using secure edge and cloud computing,’’ Inf. Sci., vol. 504, Scenarios and security issues,’’ in Proc. Federated Conf.
pp. 589–601, Dec. 2019. Comput. Sci. Inf. Syst., Sep. 2014, pp. 1–8.
[25] Y. Jararweh, A. Doulat, O. AlQudah, E. Ahmed, M. Al- [45] A. Sahai and B. Waters, ‘‘Fuzzy identity-based encryption,’’
Ayyoub, and E. Benkhelifa, ‘‘The future of mobile cloud in Proc. 24th Annu. Int. Conf. Theory Appl. Cryptograph.
computing: Integrating cloudlets and mobile edge Techn. (EUROCRYPT), 2005, pp. 457–473.
computing,’’ in Proc. 23rd Int. Conf. Telecommun. (ICT), [46] M. Blaze, G. Bleumer, and M. Strauss, ‘‘Divertible protocols
May 2016, pp. 1–5. and atomic proxy cryptography,’’ in Proc. 17th Annu. Int.
[26] M. Armbrust, A. Fox, and R. Griffith, Above the Clouds: A Conf. Theory Appl. Cryptograph. Techn. (EUROCRYPT),
Berkeley View of Cloud Computing. Berkeley, CA, USA: 1998, pp. 127–144.
EECS Dept., 2009. [47] R. L. Rivest, L. Adleman, and M. L. Dertouzos, ‘‘On data
[27] J. Ren, H. Guo, C. Xu, and Y. Zhang, ‘‘Serving at the edge: A banks and privacy homomorphisms,’’ Found. Secure Comput.,
scalable IoT architecture based on transparent computing,’’ vol. 4, pp. 169–179, Oct. 1978.
IEEE Netw., vol. 31, no. 5, pp. 96–105, Aug. 2017. [48] K. Yang and X. Jia, ‘‘Data storage auditing service in cloud
[28] H. Bangui, S. Rakrak, S. Raghay, and B. Buhnova, ‘‘Moving computing: Challenges, methods and opportunities,’’ World
to the EdgeCloud-of-things: Recent advances and future Wide Web, vol. 15, no. 4, pp. 409–428, Jul. 2012.
research directions,’’ Electronics, vol. 7, no. 11, p. 309, 2018. [49] A. N. Toosi, R. N. Calheiros, and R. Buyya, ‘‘Interconnected
[29] S. Carlini, ‘‘The drivers and benefits of edge computing,’’ in cloud computing environments: Challenges, taxonomy, and
Schneider Electric–Data Center Science Center, 2016. survey,’’ ACM Comput. Surv., vol. 47, no. 1, pp. 1–47, Jul.
[30] C. K. M. Lee, Y. Z. Huo, S. Z. Zhang, and K. K. H. Ng, 2014.
‘‘Design of a smart manufacturing system with the application [50] D. S. Touceda, J. M. S. Cámara, S. Zeadally, and M. Soriano,
of multi-access edge computing and blockchain technology,’’ ‘‘Attributebased authorization for structured Peer-to-Peer
IEEE Access, vol. 8, pp. 28659–28667, 2020. (P2P) networks,’’ Comput. Standards Interfaces, vol. 42, pp.
[31] W. Huo, F. Liu, L. Wang, Y. Jin, and L. Wang, ‘‘Research on 71–83, Nov. 2015.
distributed power distribution fault detection based on edge [51] Y. Yang, X. Zheng, X. Liu, S. Zhong, and V. Chang, ‘‘Cross-
computing,’’ IEEE Access, vol. 8, pp. 24643–24652, 2020. domain dynamic anonymous authenticated group key
[32] J. Barthélemy, N. Verstaevel, H. Forehead, and P. Perez, management with symptommatching for e-health social
‘‘Edge-computing video analytics for real-time traffic system,’’ Future Gener. Comput. Syst., vol. 84, pp. 160–176,
monitoring in a smart city,’’ Sensors, vol. 19, no. 9, p. 2048, Jul. 2018.
2048. [52] I. Khalil, A. Khreishah, and M. Azeem, ‘‘Consolidated
[33] P. Mach and Z. Becvar, ‘‘Mobile edge computing: A survey identity management system for secure mobile cloud
on architecture and computation offloading,’’ IEEE Commun. computing,’’ Comput. Netw., vol. 65, pp. 99–110, Jun. 2014.
Surveys Tuts., vol. 19, no. 3, pp. 1628–1656, 3rd Quart., 2017. [53] M. Zhang and Y. Sun, ‘‘Analysis of the development trend of
[34] C. T. Ding, J. N. Cao, L. Yang, and S. G. Wang, ‘‘Edge China’s Internet industry in 2017,’’ China Telecom, vol. 5, pp.
computing: 67–72, 2018.
Applications, state-of-the-art and challenges,’’ Zte Technol., [54] O. Makinen, ‘‘Streaming at the edge: Local service concepts
vol. 25, no. 3, pp. 2–7, 2019. utilizing mobile edge computing,’’ in Proc. 9th Int. Conf. Next
[35] A. Nadembega, A. S. Hafid, and R. Brisebois, ‘‘Mobility Gener. Mobile Appl., Services Technol., Sep. 2015, pp. 1–6.
prediction modelbased service migration procedure for follow [55] M. Huang, A. Liu, N. N. Xiong, T. Wang, and A. V.
me cloud to support QoS and QoE,’’ in Proc. IEEE Int. Conf. Vasilakos, ‘‘An effective service-oriented networking
Commun. (ICC), May 2016, pp. 1–6. management architecture for 5G-enabled

VOLUME 8, 2020 20194

 K. Cao et al.: Overview on Edge Computing Research

Internet of Things,’’ Comput. Netw., vol. 173, May 2020, Art. no. University, China, in 2017. She is currently pursuing the
107208. master’s degree with the School of Computer Technology,
[56] T. Li, M. Zhao, and K. K. L. Wong, ‘‘Machine learning based ShenyangJianzhuUniversity.Herresearchinterest is big data
code dissemination by selection of reliability mobile vehicles analysis.
in 5G networks,’’ Comput. Commun., vol. 152, pp. 109–118,
Feb. 2020.
[57] M. Chen, T. Wang, K. Ota, M. Dong, M. Zhao, and A. Liu,
‘‘Intelligent resource allocation management for vehicles
network: An A3C learning approach,’’ Comput. Commun.,
vol. 151, pp. 485–494, Feb. 2020.
[58] Mobley, R. Keith. An Introduction to Predictive Maintenance.
Amsterdam, The Netherlands: Elsevier, 2002.
[59] N. Chen, Y. Chen, S. Song, C.-T. Huang, and X. Ye, ‘‘Smart
urban surveillance using fog computing,’’ in Proc. IEEE/ACM
Symp. Edge Comput. (SEC), Oct. 2016, pp. 95–96.
[60] F. N. Iandola, S. Han, M. W. Moskewicz, K. Ashraf, W. J.
Dally, and K. Keutzer, ‘‘SqueezeNet: AlexNet-level accuracy
with 50x fewer parameters and & 0.5MB model size,’’ 2016,
arXiv:1602.07360. [Online]. Available:
http://arxiv.org/abs/1602.07360
[61] S. Y. Nikouei, Y. Chen, and T. R. Faughnan, ‘‘Smart
surveillance as an edge service for real-time human detection
and tracking,’’ in Proc. IEEE/ACM Symp. Edge Comput.
(SEC), Oct. 2018, pp. 336–337.
[62] S. Y. Nikouei, Y. Chen, S. Song, and T. R. Faughnan,
‘‘Kerman: A hybrid lightweight tracking algorithm to enable
smart surveillance as an edge service,’’ in Proc. 16th IEEE
Annu. Consum. Commun. Netw. Conf. (CCNC), Jan. 2019, pp.
1–6.
KEYAN CAO received the M.S. and
Ph.D. degrees in computer science and
technology from Northeastern
University, China, in 2009 and 2014,
respectively. She is currently an
Associate Professor with the College
of Information and Control
Engineering, Shenyang Jianzhu
University. Her current research
interests include data management,
cloud computing, and query process and optimization.

YEFAN LIU received the B.S. degree

from the Shenyang University of
Technology, China. He is currently
pursuing the master’s degree with the
School of Computer Science and
Technology, Shenyang Jianzhu
University. His current research
interest is big data processing.
GONGJIE MENG received the
bachelor’s degree in software
engineering from the Jincheng College, Nanjing University of
Aeronautics and Astronautics, in 2019. He is currently pursuing
the master’s degree in software engineering with Shenyang
Jianzhu University. He is currently conducting research on
pulmonary.

QIMENG SUN was born in Tieling, China, in 1994. She received

the B.Eng. degree from Shenyang Urban Construction

20195 VOLUME 8, 2020