Quiz 01 Ren

A Survey on End-Edge-Cloud Orchestrated Network
Computing Paradigms: Transparent Computing, Mobile

Edge Computing, Fog Computing, and Cloudlet
JU REN, DEYU ZHANG, SHIWEN HE, and YAOXUE ZHANG, Central South University, China
TAO LI, University of Florida, U.S.A.
Sending data to the cloud for analysis was a prominent trend during the past
decades, driving cloud computing
as a dominant computing paradigm. However, the dramatically increasing number of
devices and data traffic
in the Internet-of-Things (IoT) era are posing significant burdens on the capacity-
limited Internet and uncontrollable service delay. It becomes difficult to meet the
delay-sensitive and context-aware service requirements
of IoT applications by using cloud computing alone. Facing these challenges,
computing paradigms are shifting from the centralized cloud computing to
distributed edge computing. Several new computing paradigms,
including Transparent Computing, Mobile Edge Computing, Fog Computing, and
Cloudlet, have emerged to
leverage the distributed resources at network edge to provide timely and context-
aware services. By integrating end devices, edge servers, and cloud, they form a
hierarchical IoT architecture, i.e., End-Edge-Cloud
orchestrated architecture to improve the performance of IoT systems. This article
presents a comprehensive
survey of these emerging computing paradigms from the perspective of end-edge-cloud
orchestration. Specifically, we first introduce and compare the architectures and
characteristics of different computing paradigms.
Then, a comprehensive survey is presented to discuss state-of-the-art research in
terms of computation offloading, caching, security, and privacy. Finally, some
potential research directions are envisioned for fostering
continuous research efforts.
CCS Concepts: • Human-centered computing → Ubiquitous computing; • Computer systems
organization → Cloud computing;
Additional Key Words and Phrases: End-edge-cloud orchestration, network computing,
transparent computing, mobile edge computing, fog computing, cloudlet
ACM Reference format:
Ju Ren, Deyu Zhang, Shiwen He, Yaoxue Zhang, and Tao Li. 2019. A Survey on End-
Edge-Cloud Orchestrated Network Computing Paradigms: Transparent Computing, Mobile
Edge Computing, Fog Computing,
and Cloudlet. ACM Comput. Surv. 52, 6, Article 125 (October 2019), 36 pages.
https://doi.org/10.1145/3362031
This article is supported by the National Natural Science Foundation of China (No.
61702562 and 61702561), Huxiang Young
Talents Plan of Hunan Province, 111 Project (No. B18059), International Science &
Technology Cooperation Program of
China (No. 2013DFB10070), and China Hunan Provincial Science & Technology Program
(No. 2012GK4106).
Authors’ addresses: J. Ren, D. Zhang (corresponding author), S. He (corresponding
author), and Y. Zhang, School of Computer Science and Engineering, Central South
University, Changsha 410083, China; emails: {renju, zdy876, shiwen.he.cn,
zyx}@csu.edu.cn; T. Li, Department of Electrical and Computer Engineering,
University of Florida, Gainesville, FL 32611;
email: taoli@ece.ufl.edu.
Permission to make digital or hard copies of all or part of this work for personal
or classroom use is granted without fee
provided that copies are not made or distributed for profit or commercial advantage
and that copies bear this notice and
the full citation on the first page. Copyrights for components of this work owned
by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to post on
servers or to redistribute to lists, requires
prior specific permission and/or a fee. Request permissions from
permissions@acm.org.
© 2019 Association for Computing Machinery.
0360-0300/2019/10-ART125 $15.00
https://doi.org/10.1145/3362031
ACM Computing Surveys, Vol. 52, No. 6, Article 125. Publication date: October 2019.
125
125:2
J. Ren et al.
1 INTRODUCTION
In the past few decades, the development of Internet and wireless communication
technologies has
provided a very convenient channel for information exchange in people’s daily life.
By 2019, the
number of global mobile terminals increases exponentially to about 2.8 billion.
Especially with the
advance of artificial intelligence and intelligent science, the number of
intelligent lightweight devices has increased exponentially, and the
interconnection of all things has become the main trend
of the development of wireless communication networks and the Internet [125]. It
also implies the
coming of the era of Internet of Things (IoT) with a large number of sensors,
actuators, and mobile
devices deployed at the network edge. A report from Cisco shows that the monthly
global mobile
data traffic will be 49 exabytes by 2021 with a compound annual growth rate of 47%
from 2016 to
2021 [7]. A considerable part of the computing tasks generated by these devices,
such as virtual
reality, augmented reality, and industrial control, require timely and context-
aware processing.
As a result, processing massive data traffic is a key feature of the future
Internet and wireless
communication systems. Furthermore, high data rate and low delivery latency become
two key
performance indices of the future Internet and wireless communication networks. It
implies that
powerful computation devices need to process massive data traffic, and high data
rate transmission
links are also necessary to transfer the data traffic for the Internet and wireless
communication
networks, respectively.
From the perspective of wireless communication systems, ultra-dense networks,
massive
multiple-input multiple-output (MIMO), and high-frequency communications have been
regarded
as promising ways to meet the growing demands of future wireless communications,
such as
5G wireless systems. Compared to 4G, 5G is envisioned to receive a 1,000× capacity
increase by
leveraging these technologies. Besides the dramatic capacity enhancement, it is
also anticipated to
achieve significant improvement in data transmission rate, network reliability,
spectral and energy
efficiency, and so on [133]. It also implies that the future wireless communication
technologies designed in 5G systems provide powerful capabilities to convey the
data traffic generated by various
communication devices.
However, in the past few decades, various computing architectures and paradigms are
designed
to provide powerful capabilities of processing data traffics from the view point of
the Internet.
Since the advent of the first computer, ENIAC, the development of the computer, the
Internet, and
information technologies has brought people into an era of information explosion.
To further meet
the various requirements of the information society, there is a need for
revolutionary changes in
computer networks, computing modes, storage modes, and application modes. The
development
of computing modes has gone through the stages of single computer computing,
cluster computing, network computing, and cloud computing. The appearance of
cluster computing is to address
the shortcoming of single computer computing, which cannot process gigantic
computing data
services. More flexible network computing is developed to increase the business
processing capabilities of cluster computing in terms of heterogeneity, dynamics,
distribution, and scalability.
Although network computing can provide considerable computation power to process
data, it still
cannot satisfy the ever-increasing demands caused by the exponentially growing
mobile devices
and data traffic.
Cloud computing with centralized computing and storage resources, which has been
regarded
as the second generation of network computing and considered as one of the most
promising
technologies in 21th century, provides powerful capabilities of computation and
storage to address the computing challenges. In particular, cloud computing can
provide elastic services and
data-intensive analysis for end-users over a wide area network (WAN). Therefore,
users can
be empowered with seemingly unlimited resources without building new computing
infrastructures. The global revenue brought by cloud computing is forecasted by
Gartner to grow from
A Survey on End-Edge-Cloud Orchestrated Network Computing Paradigms
125:3
$209.2 billions to $246.8 billions [8]. With the remarkable economic benefits of
cloud computing,
it is likely to stay firmly on the computing landscape [129]. However, although the
centralization
of computing resource in cloud facilitates resource management and maintenance,
there are diffculties for cloud computing to satisfy the service demands of the new
trend of delay-sensitive
applications in the IoT era. The first issue is the unacceptable WAN latency, which
is unlikely to
be improved in the foreseeable future, since the design objective of WAN mainly
focuses on improving the efficiency of bandwidth and links [130]. The second issue
is that the traffic capacity of
WAN will be significantly challenged by the dramatically increasing amount of data
generated by
IoT devices. For example, in an airport surveillance application, several thousand
video cameras
are deployed for security purposes, each of which produces data at 12 Mbps [10]. To
analyze solely
video data at the central cloud server, hundreds of Gbps bandwidth is required to
collect the video
data, which far exceeds the traffic capacity of current WANs. Last, cloud computing
has intrinsic disadvantages of supporting context-aware computing for IoT
applications, since it works in a
remote and centralized computing way.
To address these issues, several new network computing paradigms have emerged to
offer computing resources in the proximity of end-users. In such a way, delay-
sensitive and context-aware
services can be offered without the involvement of WAN [90]. Emerging network
computing
paradigms, including transparent computing (TC), fog computing (Fog), mobile edge
computing
(MEC), and cloudlet, have attracted extensive attention in industry and academia.
These paradigms
employ small-scale edge servers with limited computation resources to timely serve
end-users at
the network edge. The edge servers can either be temporary devices such as smart
phones, laptops,
advanced routers, and micro servers, and so on, and also can be some nearby
infrastructures. Fog,
MEC, and cloudlet can be deemed as extending cloud services to the network edge,
since they exploit the similar computation offloading and storage management
schemes. However, in the vision
of TC, computing and storage are separated into end-devices and remote servers
[176]. Specifically, TC encourages end-devices and their nearby devices to
undertake the computing tasks and
fetch the software and data from remote servers. As a result, the computing
capabilities of modern
devices can be fully exploited. It is notable that all the above-mentioned
computing paradigms
emphasize serving end-users at the edge and to serve the delay-sensitive
applications in the IoT
context.
Motivated by the advances of the emerging computing paradigms, we are likely to see
a hierarchical computing architecture that can revolutionize the current cloud
computing architecture [50]. It consists of large-scale central servers, numerous
edge servers deployed at the network
edge, and a huge number of distributed end devices. Instead of considering them as
separated parts,
most applications require all of them to be well orchestrated for providing
reliable services over
different temporal and spatial scales. For example, in airport surveillance
applications, the edge
servers can analyse and filter the video streams before uploading the whole batch
to the central
server, which can significantly decrease the traffic flow over the WAN and relieve
the burden on
the central server without performance loss. Furthermore, considering a scenario
where an enduser requests some content of interest from an edge server offering
caching services, the central
server can work as a complementary to the edge servers with limited storage
capacity.
Several works survey cloud computing or edge computing from different perspectives.
In Reference [68], the authors discuss the architecture and performance
optimization approaches in mobile cloud computing, which employs centralized cloud
servers to offer computation offloading
and storage for mobile users through WAN. Reference [174] surveys the computing
techniques
for big data analytics, including cloud computing, TC, and fog computing. It is
believed that analyzing data at the network edge, rather than the central cloud
server, may be a better solution in
an IoT context, due to the constrained bandwidth of WAN and the context-aware
requirement. In
125:4
J. Ren et al.
Reference [113], the authors discuss the existing works on computation task
offloading in an edge
computing paradigm, mainly focusing on the tradeoff between energy consumption and
delay.
Reference [150] surveys the works of caching strategies in the context of radio
access networks
(RANs) enhanced by edge computing.
Although the above-mentioned surveys are inspiring, none of them pay specific
attention to
research issues in the hierarchical computing architecture, which provides great
benefits offered
by the orchestration of end devices, the edge, and the cloud. The scope of this
survey covers a
comparison of different network computing paradigms and different research issues,
including
computation offloading, caching, security, and privacy under the hierarchical
computing architecture. The remainder of this survey is organized as follows.
Section 2 introduces and compares
the emerging computing paradigms. Section 3 reviews the research of computation
offloading in
the emerging computing paradigm. Caching strategies and security and privacy
protection mechanisms are summarized in Sections 4 and 5, respectively. Section 6
outlines some potential future
directions in this emerging filed of study, followed by a conclusion, given in
Section 7.
2 EMERGING COMPUTING PARADIGMS AND EVALUATION CRITERIONS
In the past few years, to address massive data computation, various computing
paradigms have
been proposed to provide timely and resource-efficient services. In this section,
we introduce the
computing paradigms emerging in recent years, including transparent computing
(referred to as
TC, hereafter), fog computing (referred to as Fog), mobile edge computing (MEC),
and cloudlet.
Although they face some common issues to deal with the huge amount of computing and
storage
tasks generated by heterogeneous devices, such as the management of computing and
storage resources and networking, they also exhibit different characteristics due
to various original driving
forces, such as persuasive computing for TC and IoT applications for Fog computing.
For instance,
all these computing paradigms need to mask the heterogeneity of various devices to
ease the resource management. To this end, TC adopts the operating system–(OS)
level solution, e.g., Meta
OS [178], while the other three paradigms focus on virtualization and
containerization solutions.
2.1 Transparent Computing
Transparent computing was proposed to decouple the software, including operating
systems
(OSes), from the heterogeneous hardware of IoT devices [172]. It masks the details
of service provisioning and serves the users in a totally “transparent” way [123].
To this end, TC enables devices to
choose services on demand via networks, without considering the details of service
provisioning,
such as the upgrade and management of softwares. Generally speaking, TC can be
characterized
by the following properties:
(1) Working in a client-server mode, TC logically integrates the devices
distributed across
the network as one system. The system intelligently provides services according to
the
capabilities of devices and the conditions of the networks. The server side is in
charge of
the centralized resource management for the network-connected clients to provide
elastic
services.
(2) To logically split the software from the hardware of heterogeneous client
devices, TC
develops an on-demand service loading and execution architecture, empowering the
client
devices with the capability of dynamically executing cross-platform services from
remote
servers via high-speed networks.
(3) To fully use the computation resources residing in the client devices, TC
enables client
devices to fetch remote services from the server side on-demand and execute them
locally
in a block-streaming way. Block-streaming execution means that when users request a
125:5
Fig. 1. An illustration of the TC architecture.
Fig. 2. An illustration of TC-based IoT system.
specific service, only the necessary part of codes related to the service, rather
than the
whole software, will be loaded and executed on client devices. Consequently, the
energy
efficiency and delay of service provisioning can be significantly improved.
We describe a typical implementation of TC in Figure 1. It extends the classical
von Neumann
architecture in both spatial and temporal domains. In the architecture of TC, a
single computer
is extended to network-connected computers/devices. As shown in Figure 1, the
servers store the
system/service software that can be dynamically loaded to client devices for
execution via the Meta
OS platform. Built on an underlying OS, the Meta OS is designed to shield the
heterogeneity of
hardwares and unifies the interfaces to upper layers. In such a way, various
commodity OSes and
applications can be initialized and managed by the Meta OS platform. The Just-In-
Time Computing
layer is designed to enable the client devices to load the instructions of the
demanded programs
and user data from servers through block-streaming. After remote loading, the
client devices can
perform the computation with the local resources of client devices in a timely
manner. Under
this architecture, the storage of the client device is extended to other network-
connected devices
(or servers), and the I/O interrupts are redirected from the system bus of the
local device to the
network [173].
Recently, the advantages of TC have been further recognized in the IoT era, which
urges a welldesigned solution to manage the huge amount of software for highly
heterogeneous hardware
infrastructures, ranging from high-end servers, laptops, and smart phones to low-
end sensors. Ren
et al. [123] propose a scalable TC-based IoT architecture, as shown in Figure 2, to
provision flexible
125:6
J. Ren et al.
Fig. 3. A typical BaaS architecture.
and timely services at the edge of the network. In their proposed architecture,
nearby edge servers
act as the TC server to store some frequently used Apps and data and achieve timely
response for
the service requests of IoT devices, while the cloud acts as the storage and
management center
to provide centralized control for the whole system. A LiteTCOS is developed as the
MetaOS for
IoT devices and is responsible for commodity OS loading and block-streaming service
execution.
Moreover, He et al. [67] propose a new service model based on TC, named Block-
stream, as a
Service (BaaS) to achieve ambient service computing for IoT devices. They present a
clear BaaS
architecture, as shown in Figure 3, where any kind of software, including OS,
libraries, middleware,
and Apps, is divided into a set of code blocks. When the server receives a service
request, part of
the code blocks will be dynamically provisioned to IoT devices for timely and
efficient response.
2.2 Mobile Edge Computing
MEC is initiated by the European Telecommunication Standards Institute to enable
cloud computing services in proximity of the mobile subscribers [9, 27, 96]. By
deploying MEC servers at
the macro or micro base stations, MEC can improve the user experience by processing
the user
request at the network edge with reduced latency and location-awareness, as well as
alleviate the
load over the core network, as shown in Figure 4 [12]. Together with network
function visualization and software-defined network, MEC has been deemed as a key
enabling technology toward
the 5G era [70, 167].
To stimulate the seamless involvements of vendors, service providers, and third-
party players on
MEC, an industry standardization group has been established in ETSI to develop
specifications for a
standardized and open MEC environment. The members of ETSI MEC ISG include Huawei,
Intel,
Nokia, Vodafone, NTT DOCOMO, and so on. The first introductory technical white
paper was
published in 2014 to specify the concept of MEC and the reference architecture of
MEC platform
[1]. Furthermore, it also discusses the key enabling techniques and challenges in
MEC. During
2015 to early 2017, the group has documented several specifications of MEC, ranging
from the
terminology, service scenarios to technical requirements in MECs [2, 4, 5, 9].
Although the group
claims that MEC mainly focuses on the integration of cloud computing technology
into cellular
125:7
Fig. 4. Architecture of mobile edge computing.
networks in the early age, the name of MEC is now changed to multi-access edge
computing,
which can reflect the growing interests from non-cellular operators [3].
In recent years, numerous solutions have been proposed by both academia and
industry to enhance the performance of MEC, such as modeling, multiuser resource
allocation, and system implementation, and so on. Several surveys have focused on
the progress of MEC from different
perspectives [12]. Some real-time MEC application scenarios are discussed in
References [12, 96,
113, 139, 150]. They also discuss the taxonomy of MEC from different viewpoints,
such as the
characteristics, actors, access technologies, applications, key enablers, and so
on. A survey on the
fundamental key enabling technologies of MEC is presented in Reference [139]. It
discusses MEC
orchestration by taking both system performance and MEC platforms into
consideration, shedding light on the different orchestration deployments. In
addition, the authors also introduce the
architectures and typical deployment scenarios of MEC. The authors of Reference
[96] survey the
state-of-the-art MEC studies, focusing on the joint management of communication and
computation resources. The survey in Reference [113] discusses the key use case in
MEC, i.e., computation
offloading. The authors of Reference [150] survey the key enablers of MEC,
including cloud computing, SDN/NFV, and smart devices. Besides, they also discuss
some key technologies in mobile
edge networks, covering cloud technology, and SDN/NFV, as well as smart devices.
2.3 Fog Computing
The Fog computing paradigm was first coined by Cisco in 2012 [22]. In some sense,
fog computing
is similar with the concept of MEC. However, it is also a novel network computing
architecture
that provides the capabilities of computing at the network edge [27]. Fog computing
is originally
proposed for the context of IoTs, which demands location awareness and timely
response in addition to wireless access and mobility support. In addition, fog
computing uses an n-tier architecture
to offer more flexible services, which highlights that all the network devices
along the data routing path can provide data computing and storage services for end
devices. As shown in Figure 5,
a fog tier is physically placed between the cloud and the IoT devices to enable
compute, storage,
and networking resources pooling. The fog tier consists a large number of
heterogeneous nano
servers, ranging from dedicated devices such as edge routers, set-up boxes, and
temporary devices
such as smart phones, high-end sensors, and vehicles [21].
125:8
J. Ren et al.
Fig. 5. Architecture of fog computing.
Typical application scenarios of fog computing are the ones that require both real-
time control to
improve the instantaneous system performance and long-term batch data analytics to
gain insights
for business strategy adjustment, which calls for the interplay between fog servers
and the cloud
server. Real-time control demands the densely distributed fog servers to reduce
service latency, and
the batch data analytics naturally falls into the expertise of cloud servers with
massive computing
and storage resources. To promote the development of fog computing, the Openfog
Consortium
was founded in 2015 by the leading companies in the IT industry, including Cisco,
Intel, Microsoft,
Princeton University, Dell, ARM, and so on. Numerous use cases of fog computing can
be found
in Reference [21], ranging from smart traffic lightweight system to wind farms and
smart grid.
Much of the literature has surveyed the research issues of fog computing with
different points
of focus [27, 144]. Vaquero and Rodero-Merino [144] overview the enabling
technologies and future development of fog computing. The survey in Reference [161]
overviews the fog computing
definition and some typical application scenarios, as well as clearly presents the
challenges in fog
computing system design and implementation. The authors of Reference [160] claim
that fog computing is expected to be a natural platform for many promising and
challenging IoT scenarios. The
authors of Reference [47] describe the advantages of fog computing for IoT and
introduce various
application scenarios that are suitable for fog computing. The authors of Reference
[104] put forward 10 questions and give out the corresponding answer to demonstrate
the advantages of fog
computing compared with the other existing computing paradigms. More recently, the
authors of
Reference [27] propose a concise set of evaluation criteria in fog computing.
2.4
Cloudlet
The concept of cloudlet was first proposed in 2009 by a research team from Carnegie
Mellon University [130]. The term “cloudlet” refers to micro data centers that are
placed in the proximity
of mobile users, e.g., in a coffee shop or a classroom. The key motivation behind a
cloudlet is
to boost the interactive performance of mobile applications, especially the ones
with stringent
125:9
Fig. 6. Architecture of cloudlet-based computing [131].
requirements on end-to-end latency and jitter. Such applications demand that the
responding delay is in the order of milliseconds, which is unlikely to happen
through the Internet. To fill this gap,
the proximity of cloudlets enables the servers to provide highly responsive cloud
services to mobile users and hence complement the three-tier cloud hierarchy, i.e.,
mobile users-cloudlet-cloud,
as shown in Figure 6. In addition to provide timely service, this hierarchy
enforces the privacy of
its owners by denaturing the private data before releasing them to the cloud [131].
Several typical use cases of cloudlets have been identified, such as assisted
driving support, sports training
assistants, and so on.
After the invention of the cloudlet, the research group in CMU has published a
series of works
mainly focusing on two subjects, i.e., identifying the valuable applications for
cloudlet [129] and
designing virtual machines to support user mobility [66]. In Reference [129], the
authors state
that real-time cognitive assistance, e.g., face and speech recognition, can be the
“killer app” for
the cloudlet, which demands timely response to unobtrusively guide users’
attention. Based on an
open source ecosystem for cloud computing called OpenStack, Reference [66] proposes
the design
of a VM overlay to enable cloudlet discovery and hand-off in the mobile context,
while minimizing
data exchange among cloudlets. Due to the potential business opportunities brought
by cloudlets,
CMU and several leading industrial companies, including Intel, Nokia, Crown Castle,
Vodafone,
and Deutsche Telekom, formed the open edge computing initiative [108], aiming at
developing
key technologies surrounding cloudlets and conducting user acceptance testing.
Instead of deploying dedicated computing facilities in the proximity of users,
several existing
works utilize the idle computing resources of mobile devices to perform computing
services, which
is termed “Ad hoc cloudlet” [36]. In an ad hoc cloudlet, the mobile devices connect
with each
other through short-range radio communication technologies to perform data
analytics for IoT
devices with limited computing capabilities, such as sensors for RFIDs. Although
utilizing the idle
resources of mobile devices requires minimum efforts on facility deployment and
maintenance,
the mobility and dynamic local load of mobile devices impose significant challenges
to enable
efficient computing and service provisioning in an ad hoc cloudlet. To address this
issue, some
125:10
J. Ren et al.
Table 1. Similarities and Differences Comparison of Various Computing Paradigms
Paradigms
TC
Cloudlet
Location for Computing

End devices and nearby
devices
Base stations and
nearby devices
Devices along the
routing path
Nearby cloudlets
Cloud
Central Cloud
MEC
Fog
Virtualization
MetaOS
VM and container
VM and container
VM
VM
Research Focuses
Cross-platform and on-demand
service provisioning, and so on
Uplink computation offloading
downlink caching, and so on
Uplink computation offloading,
downlink caching, and so on
Computation offloading, VM
management, and so on
Workflow scheduling, VM
management, and so on
works design stochastic optimization approaches to improve the adaptability of ad

hoc cloudlets
to system dynamics [119].
2.5
Discussion of the Similarities and Differences
The common features of the aforementioned emerging computing paradigms are to

reduce the
complexity and cost of hardware implementation, to reduce the latency, as well as
to improve the
quality-of-experience and network efficiency. However, there are some subtle
differences that need
to be clarified, as illustrated in Table 1.
First, the motivation of the TC paradigm is to address the heterogeneities of
various end-users
in IoT networks and support different OS and cross-platform application software.
Thus, the complexity and cost of hardware implementation are reduced and the
compatibility of applications are
enhanced. On the contrary, the motivation of the introduction of fog, MEC, and
cloudlet paradigms
is to process data at the proximity of end devices to reduce the response delay and
improve user experience and network efficiency. Note that cloudlets mainly use
virtual machine for virtualization,
while MEC and fog consider containers.
Second, for the TC paradigm that emphasizes providing service-oriented computing
solutions,
the computation is executed at the end-users/edge-nodes via acquiring the OS and
application
software from the edge/cloud server. Moreover, by block-streaming service execution
method,
lightweight terminals can only execute the necessary codes rather than the whole
program of an
application to enhance the energy efficiency of service computing. For Fog, MEC,
and cloudlet
paradigms, in general, end devices and edge servers all install OS and specified
application software. Edge servers can provide timely data processing to the
service requests of end devices, while
end devices can improve their energy efficiency by offloading some computation-
intensives tasks
to the edge servers or the cloud server.
Third, from the perspective of security, these computing paradigms have different
research focuses. TC aims at leveraging the computing capabilities of end devices
and edge servers to provide
secure services for IoT devices. Security mechanisms are generally deployed at the
MetaOS layer
(both of the end device side and the edge server side) to detect malware and OS-
level attacks, because MetaOS works under the commodity OS and has the higher
priority to directly check the
program codes running in the hardware by occasionally hanging up the upper-layer
OS. But for
fog, MEC, and cloudlet paradigms, researchers mainly focus on securing the
computing environment of edge servers and offloading the security burden (such as
encryption and authentication)
from end devices to edge servers. Therefore, finding ways to leverage the
orchestration of end
125:11
devices, edge servers, and the cloud to design sophisticated security solutions
becomes a hot research trend for all the network computing paradigms [124].
2.6
Evaluation Criteria
The full benefits of the aforementioned emerging computing paradigms are leveraged
on some
common performance criteria for investigating the architecture and the optimization
algorithms
of these networks [27].
The first criterion (C1) is the need to support heterogeneity in resources. In
emerging computing paradigms, different access nodes at the edge of network and
end-users exhibit strong
heterogeneity in terms of computational and storage capabilities. Therefore, one
needs to take
the heterogeneity of various access nodes into account for designing the computing
architecture
and optimization algorithms. For example, the edge server of transparent computing
needs to
provide various OS and application software to support a variety of end-users,
which generally
have different hardware configurations. To accommodate heterogeneous services over
heterogeneous end-users, Intel provides an HTML5-based solution for achieving TC.
Compared with the
previous versions, HTML 5 possesses some new features, such as rich semantic
information and
multi-thread support. Although HTML5 possesses the advantages of cross-platform
support and
low development costs, challenges remain due to the differences between browsers
and Web security. The TC paradigm enables the implementation of browser engines at
the Meta OS layer to
shield the differences between browsers [173].
The second criterion (C2) is the stringent QoS requirements in the emerging
computing
paradigms. In future communication network, QoS requirements, such as low latency,
high data
transmission rate, high spectrum, and energy efficiency, are the key performance
indicators for
evaluating the performance of new communication and computing technologies. The key
motivation of introducing emerging computing paradigms is to reduce the data
delivery latency while
increasing the data delivery rate [49]. For example, the delivery latency needs to
be smaller than
10 ms in vehicle-to-vehicle communication or intelligent vehicular communication
network, and
the transmission rate needs to be larger than Gigabits for virtual reality.
Therefore, one needs
to take these requirements into consideration for the design of the network
architecture and the
optimization algorithms of emerging computing paradigms.
The third criterion (C3) is the need for elastic scalability. The main goal of the
emerging computing paradigms is to realize low latency, high data rate, and high
energy efficiency for a large
variety of end-users in IoT scenarios. The computing paradigms are expected to
provide services
for millions or, even more, of end-users and applications. Accordingly, they need
to have the ability
to provide an elastic on-demand service for a variety of terminals. It also means
that the architecture and the corresponding algorithms of the emerging computing
paradigms need to have the
capabilities of adapting to the changes in network scale.
The fourth criterion (C4) is whether the computing paradigm can support mobility.
For example,
in vehicular networks, the movements of vehicles significantly affect the
performance of the communication system. In particular, the fast changes of the
channel need to be carefully considered
for emerging computing applications.
The fifth criterion (C5) is the need of federation and interoperability. In the
emerging computing paradigms, the edge servers have geographically distributed
deployment on a wide scale and
are provided by different owners. The cooperation among different providers can
facilitate the
usage of various services and improve the experience of end-users. To this end, it
calls for welldesigned data and control interfaces that can enable interoperability
at different levels of providers
and architecture. For example, in a TC paradigm, Meta OS is responsible to manage
hardware and
125:12
J. Ren et al.
software resources, including OS and various application software (Apps), and to

provide computing services for users securely and reliably [172].
3
COMPUTATION OFFLOADING FOR EMERGING COMPUTING PARADIGMS
To liberate mobile devices from the limited computation capabilities and energy
supply, both industry and academia take computation offloading as a promising
solution to efficiently harmonize the computing resources in the end-edge-cloud
orchestrated computing paradigms [168]. The
resource-limited end devices, such as mobile phones, can offload the compute-
intensive tasks to
the powerful facilities to either the edge servers or remote cloud servers to
converse energy consumption and reduce the response time. In the following, we
divide the existing literature into
three categorizes according to their objectives, i.e., minimizing energy
consumption, minimizing
delay, and jointly optimizing energy consumption and delay.
3.1 Minimizing Energy Consumption
Research reports have shown that information and communication technology (ICT)
sector is responsible for 0.75 million tons of CO2 gas emissions for 1 TWh of
energy consumption. Motivated by this observation, energy-efficient computation
offloading has attracted extensive attention. Considering the energy consumption
caused by both code compilation and execution, Chen
et al. design an offloading scheme to strike the balance between computation and
communication
in Reference [33]. A partial offloading scheme is proposed to conserve energy of
each MD for context interactive applications, taking into account the social
relationships between users [25]. In
Reference [46], the authors present an energy-aware offloading approach to enable
fine-grained
code offloading while bringing minimal burden on the programmers. This work
determines the
decision of computation offloading at runtime. The above surveyed works focus on
offloading to a
single cloud server. In Reference [107], the authors propose an energy-efficient
multisite offloading approach, in which the partitioned applications can be
executed by either the MDs or several
servers. The application is modeled as a weight object relation graph, in which a
edge weight represents the communication energy cost, and a node weight indicates
computation energy cost.
The objective is to optimize the energy consumption. In Reference [24], the authors
provide a
strategy to reduce the overall energy consumption without sacrificing the network
performance.
In Reference [153], the authors involve a small cell cloud manager into the edge
computing architecture that manages the computing-related activities of the femto-
cloud. The authors jointly
optimize the allocation of communication and computation resources based on partial
offloading.
Moreover, dynamic voltage scaling technology is used to adjust the computational
speed of MDs
to reduce energy cost or reduce execution time.
Numerous works take the dynamics in wireless connections between MDs and servers,
and the
available computation resources in servers into consideration. In Reference [54],
the authors propose an offloading decision framework to choose an optimal resource
provider, such as a local MD,
a cloudlet, or the remote cloud. The framework consists of a profiler, context
manager, offloading
decision maker, execution planner, and distributed service execution engine. The
profiler analyses
the characteristics of the operations and resource consumption profiles, based on
the context information gathered by the context manager. The execution planner
studies possible computation
policies based on data locations and context information, while the decision maker
chooses the
optimal scheduling policy and resource provider. Reference [35] finds the optimal
service mode by
the cooperation among cloudlets and remote cloud. It takes the mobility of
cloudlets into account
in the cooperative service provisioning. For example, co-located clouds-based
service can provide
fine-grained mobility support at the cost of potential lower computing capability
in comparison to
the powerful remote cloud. The authors schedule the computation offloading in an
opportunistic
125:13
way to enable high mobility while minimizing cost. Different from traditional
service modes of remote cloud and cloudlets, the proposed method can achieve a
flexible tradeoff between energy cost
and mobility support. The optimal option of task schedule is to minimize energy
cost. Differently
from Reference [35], Reference [54] organizes MDs into clusters to provide
cloudlike services.
In Reference [26], the authors consider mobile computing offloading for
heterogeneous networks. This article supposes that there are multiple offloadable
components in one application,
each of which has different size and computation complexity. To conserve the energy
of the MD
under a given delay constraint, it presents a combinatorial optimization algorithm
to make offloading decisions. The delay includes both communication delay and
execution delay. The optimal algorithm can get around 43% energy savings. In
Reference [48], the authors consider the scenario
where small cells that equipped with computing and storage resources connect MDs
through a
high-rate wireless channel. Mobile applications can be partitioned into several
components and
some of these components can be offloaded to the small cells. Differently from
Reference [26], when
making the offloading decision, the authors take the dependency relationship among
components
into consideration and formulate it as a generic graph. Then, they propose to
minimize the energy
cost of MDs with a strict delay requirement, including communication delay and
execution delay.
In Reference [169], the authors consider 5G heterogeneous networks, including a set
of MDs and a
macro base station built-in with an edge server. They present a multi-user
computation offloading
framework to minimize the system energy cost with the delay constraints by jointly
optimizing the
offloading decision and the radio resource. Reference [65] investigates the
computation offloading in a three-level hierarchy, in which the MDs are connected
to edge servers through wireless
links, and the edge servers connect with the cloud server by optical fibers. To
minimize the energy
consumption of MDs while maintaining the delay under a threshold, the authors
propose both a
centralized optimization scheme and a distributed offloading scheme based on game
theory.
In addition, adaptive power control and renewable energy exploitation are powerful
methods
to conserve energy consumption. In Reference [93], the authors propose a
distributed power control algorithm by fine-tuning transmission power for the small
cell base stations. Consequently, it
efficiently improves the delivery ratio to end users within required delay. In
Reference [97], the authors consider a MEC system consisting of a MD equipped with
an energy harvesting component
and a MEC server. The energy cost minimization problem, which incorporates both the
execution delay and offloading failure, is formulated as a high-dimensional Markov
decision problem.
An online Lyapunov optimization algorithm is proposed to jointly decides the CPU
frequencies,
the transmit power, and the offloading decision for computation offloading. The
simulation results
demonstrate that the proposed algorithm can significantly improve the performance
in terms of
energy consumption and effectively decreases offloading failure. In Reference [40],
the authors
consider a three-layer computing architecture, consisting of one MD, one edge
server at the edge,
and one remote cloud server. The edge server decides whether to process the tasks
or further offload it to remote clouds, with the objective to minimize the weighted
sum of energy consumption
and delay. To address this problem, an efficient heuristic algorithm are proposed
by using semidefinite relaxation and a randomization mapping approach. Simulation
results illustrate that the
proposed approach enables substantial improvements via using a computing access
point between
the remote cloud and the MD.
References [97, 165] consider the cases where the MDs can be powered by energy
harvested
from the ambient energy sources to further prolong their lifetime. In Reference
[97], Mao et al.
schedule an energy harvesting–(EH) powered MD to offload its task to a proximate
edge server.
Both the CPU speed and transmission power are optimized to minimize the long-term
cost that
combines the execution delay and the penalty caused by task dropping. In addition
to EH-powered
MDs, Zhang et al. [165] consider that the execution of a task may exceed one time
slot, which leads
125:14
J. Ren et al.
to the coupling of task processing across slots. To address this issue, Reference
[165] divides the
original tasks into smaller subtasks that can be processed in each time slot.
3.2
Minimizing Delay
One of the main targets of end-edge-cloud orchestrated computing paradigms is to

reduce the
service delay. In Reference [44], the authors present an approach to optimize the
total execution
time of an application composed of multiple modules. This work introduces a dynamic
application partitioning mechanism between the user equipment and the cloud. In
Reference [59], the
authors model the execution of applications as graphs and determine the optimal
distribution of
the application between servers and MDs. First, they abstract the behavior of
application modules
as a dataflow graph. Each module offers a set of services and connects with each
other. By offline
profiling, the dependencies among modules are characterized by their resource cost
and hence can
provide some a priori knowledge for optimization. Second, a partitioning algorithm
is performed
to minimize the interaction delay. In Reference [159], the authors investigate
multi-user application partitioning problem and further consider how to partition
jointly computations of multiple
users. Instead of minimizing the service completion time for each user, the work
aims to minimize
mean completion time for all users.
In Reference [86], the authors consider the computation offloading problem in an
MEC system
comprising an MD and an MEC server. The authors find that the execution time of
most mobile
applications is in the range of tens of milliseconds, while the typical duration
value of a channel
block is a few milliseconds. It implies that the application execution process may
be across many
channel blocks, transforming the computation offloading problem as a two-timescale
stochastic
optimization problem. Specifically, the offloading decision is made in the large
timescale, and the
transmission policy is made in small timescale by taking the instantaneous wireless
channel conditions into consideration. Based on such insights, the work analyses
the average delay of each task
and proposes an efficient algorithm to minimize the offloading delay. To minimize
the mean delay
of general traffic flows in the LTE downlink, in Reference [57], the authors
introduce a mobile edge
scheduler for MEC paradigm. It is deployed closely to the eNodeB and implemented
with a channelaware and flow-aware scheduling policy. By accommodating the
transmissions to the available
channel quality of MDs, it can minimize the mean delay for the complete set of
traffic flows.
Due to several stochastic factors, such as the changing wireless connections,
fluctuation of transmission bandwidth, and user mobility, data transmission between
the mobile user and the cloud
is highly unreliable. In Reference [134], the authors demonstrate that the “bad”
connectivity consumes a lot of energy at MDs. To cope with this problem, they
propose a Lyapunov optimization to
optimize response delay, only requiring prior system knowledge. It adaptively
utilizes the duration
of good connectivity to prefetch frequently used data while deferring delay-
insensitive data in bad
communication status. Generally, users have different service delay and energy
consumption requirements for various applications. In Reference [152], the authors
design an optimization method
to minimize the weighted sum of energy consumption and the computation delay. The
proposed
method consists of two steps. First, they determine whether it should be offloaded
or not. Second, it should be offloaded to a particular server. Reference [88]
leverages the social relationships
between MDs to design a Nash equilibrium-based solution. The objective of Reference
[88] is to
minimize the social group execution cost, which is defined as the sum of the
execution delay and
the punishment caused by task dropping. Reference [82] designs a deep reinforcement
learning–
(DRL) based algorithm to deal with the instability in a device-to-device offloading
scenario. The
DRL-based algorithm takes the length of task queues at MDs and edge servers, and
their distances
as the input, and outputs the computation offloading decisions to the user utility
obtained by task
execution, while maintaining the energy consumption and delay under given
thresholds.
125:15
The aforementioned works focus on how to offload the application workloads from MDs
to
edge clouds. However, how to select an optimal edge site in the edge network to
undertake the
workloads is also critical to optimize the response delay of mobile applications.
For example, users
may conglomerate close to a single cloudlet but far away from another. However,
this situation
makes an overloading usage of the first cloudlets resources, while resulting in a
wasted capacity of
the second. For this scenario, in Reference [127], the authors propose an approach
for minimizing
offloading delay with two cloudlet servers. The article improves not only the
computation delay
by VM migration but also the communication delay by transmission power control. In
Reference
[138], the authors further explore this problem and considered the scenario that
MDs offload their
application to geographically distributed cloudlets. To achieve a minimization of
the average response time, they propose a delay-aware task offloading strategy to
allocate MDs’ computation
tasks into optimal cloudlets. The cost is both considered the network delay and the
cloudlet delay.
The remote cloud can cooperate with the edge to achieve better delay performance.
In Reference
[29], the authors consider three types of actors consisting of MDs, a cloud edge,
and a data center
cloud, to comprise a resource pool. This article exploits the Lyapunov optimization
approach to
manage resource pools for improving the overall experienced delay of mobile users.
Similarly to
Reference [29], in Reference [179], the authors consider a scenario with multi-
users, one resource
constrained local cloud and one resource-abundant remote cloud. According to the
requirements
of applications in terms of delay, the authors design a priority queue-based
threshold policy to
maximize the probability of completing application in time. Numerical results
demonstrate that
the quality of service is vastly improved with the cooperation of remote cloud when
the task
queue of edge cloud is exceeded. In Reference [89], cloudlets can determine whether
to respond
user service requests locally or offload them to the remote cloud. The authors
propose a multiresource allocation strategy to improve the QoS. Moreover, to
maximize the long-term reward
under service delay requirements, they formulate a semi-Markov decision process
problem and
address it by linear programming.
Usually, increasing workload leads to an increasing delay. To reduce the delay, the
user may
eventually increase the transmission rate at the expensive of increasing radiated
power. Accordingly, the system workload needs to make an endeavour on the tradeoff
between the delay and the
energy consumption. In Reference [105], the authors consider a system with multiple
users served
by a small cell node with computational capabilities, and partial instructions of
an application can
be offloaded to the small cell node. The authors investigate the delay and energy
consumption
tradeoff through adjusting uplink data transmission rates. Their simulation results
demonstrate
that, for a given energy cost, the delay increases with increasing number of users.
As a result,
the uplink rate should be increased to achieve certain quality-of-experience by
sacrificing energy
efficiency. Reference [98] considers the joint optimization of delay and energy
consumption during the computation process in a multiuser scenario. An online
algorithm is designed to decide
whether local execution or computation offloading, with respect to the constraints
on task buffer
stability. The proposed algorithm determines the optimal CPU frequencies of MDs in
each time
slot and the optimal transmission power and bandwidth using the Gauss-Seidel
method.
3.3
Joint Optimization of Energy Consumption and Delay
To improve the performance of service provisioning in the end-edge-cloud

orchestrated computing paradigms, some literature investigates computation
offloading by taking both the energy consumption and delay into consideration. In
Reference [84], the authors propose a task scheduling
algorithm that jointly schedules the cloud resources and wireless channels to
minimize the energy consumption with constraints on completion time. Lin et al.
describe an application as a
directed acyclic graph, in which a node and an edge represent a task and the
execution sequence,
125:16
J. Ren et al.
respectively. In Reference [171], the authors partition an application into a

sequence of tasks, aiming to optimize the energy consumption of the MD under a
execution deadline constraint. They
further study collaborative execution between the MD and the cloud by formulating
the execution process as a constrained shortest path problem. In Reference [94],
the authors minimize the
energy consumption of MDs under the constraints of execution delay and component
precedence
ordering. They design a wireless connection-aware offloading algorithm for multi-
component applications to decide the offloaded components of the application.
In Reference [180], the authors consider the scenario consisting of multiple
femtoclouds with
computation and storage capabilities and MDs, each of which has one task to
execute. The application can be partitioned into two parts, in which one part is
executed locally and the other
is executed at the edge. Targeting at minimizing the overall energy cost of MDs
under delay constraints, this work jointly optimizes the allocation of radio and
computational resources. Typically,
the proposed approach can reduce 40% average total energy when compared to the no-
offloading
solution, and the time complexity of the proposed approach is only O(K ). In
References [162,
163], the authors also consider a multiuser system that consists multiple single-
antenna MDs and
a single-antenna base station. The computation offloading problem is formulated as
a convex optimization problem to minimize the weighted sum of multiusers’ energy
consumption under delay
constraints.
In Reference [85], the authors define a customizable cost model, which enables
users to adjust
the weight of delay and energy in the optimization problem. In Reference [43], the
authors partition and offload applications at runtime to optimize the delay and
energy consumption. Similarly
to Reference [43], Reference [80] proposes a framework called ThinkAir, which
utilize the smartphone virtualization technique to facilitate the application
mitigation from smart phone to cloud.
Reference [20] considers the case where the MDs are powered by wireless energy
transferred from
the edge servers; the authors schedule the time allocation between energy transfer
and computation task offloading to maximize the computation rate, i.e., the ratio
between processed data and
the processing time. To account for the the selfishness of MDs, Reference [76]
designs a game
theory–based scheme to jointly optimize the energy consumption and delay of
computation offloading. Furthermore, the authors analyze the price of anarchy of
the scheme, which quantifies
the gap between the proposed scheme and the optimal solution.
Pu et al. [119] investigate energy-efficient computation task offloading among
multiple MDs
through cellular and D2D links. The authors consider the stochastic task arrivals
and channel conditions over time and propose an online algorithm to dynamically
offload the tasks to minimize
the long-term energy consumption. References [34, 99, 155] design online algorithms
to strike the
balance between energy consumption and execution delay. Considering an MEC system
consisting
of multiple users and edge servers, Mao et al. [99] minimize the long-term average
weighted sum
power consumption of the MEC system by adjusting the radio and computation
resources at both
the MDs and edge servers. Chen et al. [34] investigates the computation offloading
among the edge
servers deployed at small-cell base stations. The problem formulation takes the
long-term delay
cost and energy consumption of edge servers as the objective function and
constraints, respectively. Wu et al. [155] investigate the computation offloading
in a three-level hierarchy consisting
of MDs, edge server, and a remote cloud server. Considering the different bandwidth
and computation resources among the three levels, the authors propose an online
algorithm to minimize
the average energy consumption of MDs while ensuring the response time under a
given time
constraint.
In multiuser scenarios, one critical factor that affects the performance of
computation offloading is the wireless access efficiency. If too many users select
the same wireless channel to connect
the edge access node simultaneously, then they may suffer severe interference.
Reference [41]
125:17
investigates the computation offloading in an interference environment. First, the

authors model
the offloading overhead with a tradeoff factor, which denotes the weighting
parameter of computational time and energy. Then, they formulate the multiuser
offloading decision problem as a
game and propose to achieve the Nash equilibrium. Based on Reference [40], the
authors of Reference [39] further extend the study to multi-user scenario.
Different from Reference [40], this work
aims to jointly optimize communication and processing resource allocation among
competing
MDs. However, References [40, 179] only consider the offloading policy from MDs to
one server. In
Reference [63], the authors consider a heterogeneous multi-site offloading
environment, consisting
of MDs, cloudlets, and public clouds. The cloudlets act as distributed proximal
devices, such as WiFi
access points, Femtocell access points, and Macro-cell access points. These devices
might be heterogeneous in terms of processing speed, communication delay,
disconnection probability, and so
on. Public clouds are assumed to have similar transmission bandwidth and processing
speed. Tasks
are offloaded to these heterogeneous clouds or cloudlets. The authors propose a
multi-site computation offloading algorithm to achieve an optimal offloading site
for each task to minimize the
overall cost of energy consumption and execution time. It is noteworthy that fog
radio access network is the evolved network by equipping the RRHs with caching and
signal processing facilities
to improve the spectral efficiency and reduce the delivery delay between the
baseband units and
the RRHs. The work in Reference [111] optimizes the delivery phase involving both
fronthaul communications and wireless transmissions by considering channel
precoding and cached content.
4
CACHING FOR EMERGING COMPUTING PARADIGMS
In the emerging end-edge-cloud orchestrated architecture, nodes at different levels

of network
have different abilities to compute and store data files. At the same time,
different caching strategies have different impacts on the system performance, such
as the network throughput, system
energy efficiency, delivery latency, and so on. However, the dynamic
characteristics of the user behaviours, user demands, and network environments make
the caching decision very challenging.
Consequently, the research on the cache placement, contents, and strategies has
received numerous attention from both academia and industry for the end-edge-cloud
orchestrated architecture.
4.1
Placing Caching Units
In the end-edge-cloud orchestrated architecture, edge devices should be able to

provide various
application services and cache the relevant data for computation-intensive end-
users. To give full
play to the abilities of edge devices, the cache placement should be related to the
capabilities and
locations of edge devices as well as the target severed [17, 30, 38, 73].
For a transparent computing paradigm, the distributed clients need to load the
image of OS and
software stored on the centralized transparent severs to perform the computing task
via the network. A great deal of requests from distributed clients for fetching OS
and application software
results in the bottleneck of access to servers. Therefore, the transmission delay
between servers
and clients in transparent computing should be carefully considered to improve the
system performance. To address the problem, Reference [58] designs a two-level
cache structure composing
of block caches of servers and clients. The authors exploit client cache to store
the service requests
from the clients and responded data from remote servers for reducing the I/O time.
Meanwhile,
cache is also added on service handlers at the server side and two different
caching strategies are
developed for the clients and servers, respectively. The classical least-recently-
used algorithm is
employed at the client side, and a newly designed caching algorithm leveraging
frequency-based
multi-priority queues is tailored for the server side. Research shows that the
transmission delay
and the system response times can be significantly reduced by using the multi-level
cache hierarchy for transparent computing. Based on that, an improved two-level
cooperative caching strategy
125:18
J. Ren et al.
has recently proposed for transparent computing [136]. To evaluate the caching
effectiveness and
efficiency in transparent computing, a simulation framework is designed, by which
the performance of multi-level cache strategies can be evaluated according to
different cache configurations
and replacement strategies [87].
Since the essential idea of transparent computing is to extend von Neumann
architecture in
network environments [177], the performance and stability of computing depend on a
reliable
communication network. With the development of mobile communication technologies,
mobile
computing devices become more and more prevailing. Therefore, it is necessary to
investigate
novel caching technologies suited to transparent computing in mobile network
environments. In
Reference [141], the authors design a block-level cache scheme according to the
temporal feature of
access to the server and combine with local storage access technology to optimize
the block-level
caching strategy by considering the limited bandwidth and communication stabilities
of wireless
networks.
In the mobile networks based on the all-IP cellular, the capacity of the
traditional centralized
mobile network cannot satisfy the demand for the explosive growth of rich
multimedia contents.
Recently, there are some works focusing on the issues of cache placement, including
the core network, radio access network (RAN), and user devices [151]. The core
network undertakes enormous
traffic from RAN and user devices and exchanges inter- and intra-ISPs (Internet
service providers).
Therefore, it is very difficult for the core network to support massive traffic
transmission. In this
case, the core network is in the place where caching content is widely deployed.
The research efforts on caching popular content in mobile core network have proved
that the content traffic can
be reduced by one third to two thirds (Reference [55] and Reference [121]). In the
core network,
the deployment of caching contents with the evolved packet core (EPC) draws more
attention
from research. For example, References [55, 154] investigate the deployed places of
caching using
content delivery technologies.
The backhaul links between RAN and the core network and the wireless networks
between
RAN and user devices become another bottleneck for mobile networks. In
heterogeneous mobile
networks, the multilayer architecture composing of macro-cell base stations (MBSs)
and small cell
base stations (SBSs) for caching and delivering content has been regarded as
promising solutions
for transmitting multimedia traffic [132]. MBSs need to support services for more
user devices
with more coverage areas, and then they can achieve a higher hit rate by data
caching. SBSs with
limited storage capability are more close to users in comparison with MBSs, which
will be densely
deployed in 5G networks [114]. Hence, user devices can retrieve contents directly
from MBSs or
SBSs rather than from the centric cloud. Since MBSs and SBSs are in close proximity
to users, compared to the core network, the deployment of caches in the two places
can reduce traffic exchange
between the core network and clouds as well as traffic exchange between inter- and
intra-ISPs.
Therefore, some works [11, 16, 61, 116] focus on the deployment of MBS caching and
SBS caching.
For instance, Reference [11] propose a video-aware scheduling technique by using
MBS caching
to improve the video capacity and user experience. Besides, the framework relevant
only to content caching at the base stations is proposed in Reference [16], which
quantitatively evaluates
the performance of clusters of cooperative base station caching. The device-to-
device communication possesses a great many advantages, including the increase of
network spectrum efficiency,
offloading traffic from MBSs or SBSs, and the reduction of transmission delay [15,
137].
4.2 Contents of Cache Units
Caching at the network edge aims to alleviate the congestion of transmission
between edge devices and backhaul links or clouds and to balance the tradeoff among
content delivery, transmission delay, and energy consumption. Currently, increasing
multimedia services are generating
125:19
tremendous traffic on Internet. Moreover, a large number of users in mobile

networks accesses
the multimedia services by using mobile smart devices, which results in a great
challenge for the
multimedia content delivery and user experience [166].
It is worth noting that not all network traffic is suitable for being cached, since
some un-reusable
information has no need to be cached. References [121, 154] have indicated that a
majority of multimedia content traffic is attributed to the duplication
transmission of popular contents, while a
large part of users requests for accessing to only a small portion of popular
contents. Hence, existing studies focus on how to exploit the redundant request
from users for reducing the duplication
transmission between edge devices and the burden on the backhaul links.
From modelling popular contents perspective, the independence reference model
adopted for
performance analysis of web caching is utilized for most of the researches on
mobile caching. In
the independence reference model, contents are requested according to an
independent Poisson
process with the rate corresponding to the content popularity [112]. Nevertheless,
the independence reference model assumes that the popularity of contents is static
without considering the
spatial and temporal change. In edge computing, the static model of content
popularity cannot
reflect the real popularity of contents due to a great deal of heterogeneity and
extensive distribution of computing devices with mobility. Therefore, the dynamic
analysis methods attract more
attention of researchers. For instance, in Reference [28], the statistical
properties of popularity
distributions of requesting contents are analyzed and the opportunities utilizing
the latent request
for “the Long Tail” potential are discussed. The authors of Reference [142] propose
a traffic model
to capture the dynamics popularity of the contents requested by users, whereas
these works do not
consider human-driven information related to proximity to people, such as
experience and preference of users as well as locations [95]. In the periphery of
the networks, the information about the
end-users’ locations and environmental common interests should be utilized due to
the dense geographical distribution of the edge devices and proximity to humans.
Furthermore, in a different
temporal region, the demand of users is different, namely, the popularity of the
requested content may change at different temporal region. In other words, the
temporal popularity of contents
should be considered for caching.
4.3
Caching Strategies
It is crucial to decide what caching strategies to be adopted for caching at the

edge of networks. Various caching strategies have been proposed for mobile
networks, of which a portion are extended
from the traditional caching strategies in wired networks by being tailored for
mobile networks.
4.3.1 Consideration in Caching. Although some conventional caching strategies with
considering the content in cache units, such as the least frequently used, the
least recently used, and the
first-in–first-out, can alleviate the traffic congestion of transmission data with
uniform size, they
are not efficiently applied to the case with the variation of transmission data and
delay [72]. In addition, the cache size of radio access node in mobile networks is
very limited, and caching contents
in radio access network are varied over time, which results in the reduction of the
cache hit ratio. Considering the limited caching space in mobile cell networks,
caching data at an edge server
should be adjusted according to the popularity of the contents to achieve better
caching efficiency.
The authors of Reference [64] investigate the cache replacement strategy, which is
modelled as a
Markov decision process to minimize the transmission cost between edge servers in
cellular networks. Then, they calculate the traffic cost for the actions of
possible cache replacement in term
of the previous data request and traffic between edge servers.
Recently, a few works have been written to provide a survey of caching strategies
in
information-centric networking [72, 170]. The emergence of information-centric
networking
125:20
J. Ren et al.
facilitates the context-aware delivery by the deployment of in-network caching

[157]. In
information-centric networking, multiple superiorities have been verified by
leveraging innetwork caching, which includes the reduction of network traffic and
the user access latency as
well as the alleviation of server bottleneck. Specifically, in an IoT network, low-
rate monitor and
measurement data generated by large number of devices are delivered to edge servers
for preprocessing. Since the IoT data possess a transient nature, it is a main
challenge to cache content data at
routers. Currently, the method of utilizing the in-network caching technology to
mitigate the data
traffic of IoT has attracted considerable attention from the research community.
Reference [147]
studies in-network caching related to the transient content data at routers based
on a temporal
data property, referred to as the data item lifetime. It proposes an analytical
model that can well
capture the balance between the multi-hop communication cost and data item
freshness.
4.3.2 Proactive Caching Strategies. For emerging computing paradigms, computing
services are
pushed away from centralized nodes to the network edge. The perception of the
specific locations
and the common interests of end-users is benefited from the proximity of the edge
devices. Consequently, the research efforts on content caching and delivery in the
edge devices begin to consider
popularity-driven caching.
Proactive caching strategies leveraging the proximity of edge computing devices
have attracted
large number of attentions from academia and industry. In Reference [18], the
authors propose to
cache proactively the popular files at both base stations and end devices during
off-peak period
and to leverage the correlations among social networks and D2D communication links
to reduce
the peak traffic demands. The similar proactive caching strategies in cloud radio
access networks
are proposed in Reference [37]. In addition, the popularity of contents varies with
the spatial and
temporal differences, and different people have different preferences on various
files. The authors
of Reference [11] propose a proactive caching strategy based on user preference
profiles of active
users in a specific cell to maximize the count of concurrent video sessions while
matching the
initial delay demand of the end-to-end network. Besides, since edge devices with
computing and
storage capabilities are close to users, many research works explore proactive
caching by using
machine learning to track and estimate the content timely request from end-users.
In this regard,
mobile operators can benefit from the exploitation of big data analysis and machine
learning for
the content popularity due to advantages of upcoming 5G networks. Reference [164]
proposes a
big data-enable architecture, in which the estimation of content popularity can be
obtained by the
approach of big data analysis and is used to cache contents at the base stations.
Moreover, the user mobility is the most important feature of mobile networks.
Therefore, proactive caching strategies with mobility awareness have been
investigated for edge caching communication networks. Reference [149] proposes a
general architecture, in which the key properties of
user mobility are identified to address the problem of content caching in content-
centric wireless
networks. Mobility-aware methodologies are further developed to model the spatial
and temporal
properties of mobility patterns. A mobility prediction algorithm [83] is designed
to provide seamlessly content service for the users whose mobility patterns are
unknown in advance. In addition,
Gomes et al. [62] propose an architecture to enhance the migration of content-
caches located at the
edge networks by leveraging the combination of information-centric networking and
the mobile
follow-me cloud approach in 5G networks.
4.3.3 Clustered and Cooperative Caching Strategies. The key features of the edge
computing
devices include the extensive distribution and proximity to users. Hence, it is
necessary to exploit
these features to develop efficient content caching and distribution techniques for
improving the
content delivery efficiency in 5G networks.
125:21
A lot of existing clustered caching strategies aims to improve the QoS and energy
efficiency of
wireless networks from the caching content clustering perspective. To address the
problem that
the delivery of the content objects in edge caches is not accomplished between
remote radio heads
and the users in C-RANs, a cluster caching structure in the network edge is
proposed via utilizing
the centralized signal processing and distributed edge caching [181]. A similar
problem is also
solved by other cluster content caching paradigms [32, 74].
The cooperative caching strategy is another attractive topic in emerging computing
paradigms.
To transmit video contents, the authors of Reference [23] develop a lightweight
algorithm to
achieve both the maximum of traffic volume served from cache and the minimum of the
bandwidth cost for the cooperative cache management network. The authors of
Reference [75] present
a strategy of content caching and delivery to coordinately alleviate the burden on
backhaul and improve the content delivery efficiency by formulating the problems of
cooperative content caching
and content delivery as an integer-linear programming and unbalanced assignment
problem. Besides, by considering the bandwidth limitation of the base stations in
mobile networks, the authors
of Reference [117] design a joint optimization of the caching and routing problem
to maximize the
content request served by the small cell base stations.
In addition, the delivery of the multimedia content depends on a reliable
communication network [156, 158]. However, the spectrum resource increasingly
congests in mobile networks, which
limits the peak rate of the traditional cell architectures. Recently, D2D
communication becomes a
promising way to mitigate significantly the bottleneck of spectrum scarcity in
mobile networks.
Therefore, many works [60, 73] focus on the offloading of the content delivery
traffic and the improvement of the network throughput by studying cache-enabled D2D
communication. Different
from the previous work, the authors of Reference [31] consider the interference
among D2D links
due to the proximity of distance between a user and the undesired transmitters.
They propose a
cooperation strategy to resist the interference to the D2D transmission by
exploiting the caching
capability of users devices.
5 SECURITY AND PRIVACY FOR EMERGING COMPUTING PARADIGMS
Security and privacy are two key factors for promoting the flourish of network
computing applications. The orchestration of end, edge, and cloud makes the
computing paradigms not only face
the challenges inherited from cloud computing but also introduce some new
challenges, such as
device heterogeneity, supporting mobility, location-awareness, and low-latency
services. Existing
works on protecting the security and privacy for emerging computing paradigms can
be briefly
divided into two categories. The first is how to protect the security and privacy
of the computing
system itself. The second is to protect the security and privacy of the services
under the computing
paradigms, including service provisioning, data processing, data transmission, and
data storage.
In this section, the existing works are summarized from two aspects.
5.1 System-level Security and Privacy
From the security and privacy perspective, the hierarchical framework of end-edge-
cloud orchestrated computing paradigms is a double-edged sword: It provides a
powerful networking and computing architecture for security protection; however, it
also makes the different layers of the framework vulnerable to various system-level
attacks. In the following, we mainly introduce some existing works that focus on
the security issues on virtualization and how to leverage the hierarchical
framework for resisting threats and attacks, as well as intrusion detection.
5.1.1 Virtualization. For the emerging end-edge-cloud orchestrated computing
paradigms, visualization is a key technology that can create virtual machines (VMs)
to share the computation
125:22
J. Ren et al.
and storage resources of physical servers. However, since VMs are also vulnerable
to malicious
attacks and physical servers may experience system failures, it would consequently
lead to unavailability of services and resources.
To increase the system dependability and robustness when a computing node has a
system failure or encounters an attack, the authors of Reference [109] design a
smart pre-copy live migration
approach to estimate the downtime after each iteration for determining whether to
proceed to
the stop-and-copy stage or not. Since users usually adopt different terminals for
service requesting, a user-specific trusted virtual approach is designed to adapt
different security applications by
dynamically instantiating in a secure place at the network edge [102]. Meanwhile, a
user-defined
security model is proposed to ease the security configurations transparently to
heterogeneous devices. In Reference [146], the design and implementation of self-
configuring honey-pots is designed
to adapt the observed environment. Since in-trusted IoT devices might be
interconnected toward
the aggregation networks and external malware could be applied to the network,
security issues
may arise. The authors of Reference [145] propose edge-to-edge IoT security
architecture based on
network function virtualization for monitoring the current flows in IoT systems and
identifying
malicious flows through different anomaly detection mechanisms.
5.1.2 Threats and Attacks. To reduce the threats and attacks, some researchers
intend to reinforce IoT systems from the framework perspective. By using both the
centralized controller and
distributed controllers, a software-defined network-based master-slave security
architecture is designed in Reference [77] to enhance the security and privacy
protection for the cloud users by
reducing the communication distance. Moreover, an attack pattern signature on the
web interface
in security agent unit is defined in Reference [14] to detect persistent threats in
distributed data
center network and to the module on floodlight for detecting and blocking hosts
with the traffic
pattern in all data centers. A broad class of attacks can be detected via this
method such that the
security of data centers is enhanced.
5.1.3 Intrusion and Malware Detection. Intrusion and malware detection in the cloud
environment has been studied for many years. However, these kinds of solutions
cannot work well for the
lightweight edge servers and end devices. It consequently attracts increasing
research focuses on
designing tailored intrusion detection mechanisms for end-edge-cloud orchestrated
systems. The
authors of Reference [69] develop a lightweight and distributed intrusion detection
system. It is
distributed in a three-layered IoT structure, where the cloud layer is to cluster
primary network
traffic and train its detectors, the fog layer is to analyse intrusion alerts, and
the edge layer is to
deploy detectors. A more general threat model is proposed in Reference [110] to
design the security protection method. To detect the intrusion in private cloud,
Rajendran et al. [120] propose a
hybrid intrusion detection algorithm through incoming request scanning.
The detection of malware and intrusion in TC is based on its specific architecture,
which provides some new clues for security mechanism design. In Reference [178],
Zhang et al. propose a
Meta OS platform to manage the hardware and software resources of a networked
system, with the
objective of providing secure and reliable computing service for end users. The
Meta OS, which
locates between the hardware and the OS layer, can be used to enhance the security
of TC via
monitoring the software and user data. A remote booting integrity service mechanism
[81] is further designed to resist attacks in TC. In addition, through the
streaming service execution of TC,
all data obtained from the server will be eliminated from end-users after
execution, and, thus,
viruses have few opportunities to jeopardize those terminals. Both end devices and
edge servers
can monitor and manage the data streams during the transmission process [123, 175].
125:23
5.2 Service Security and Privacy

One of the biggest security challenges in IoT systems is that lightweight end
devices usually have
insufficient resources (e.g., computing and storage) to deploy traditional security
protection mechanisms. The end-edge-cloud orchestrated computing paradigms enable
edge servers and the cloud
to take the security burden of end devices and build a more secure IoT system. For
example, we
can leverage edge servers to perform local security monitoring, threat detection,
and threat protection on behalf of the endpoints [42]. However, when we consider
the services provided by these
computing paradigms, we should think about the following questions: whether the
devices should
be trusted; how to preserve the data privacy when edge servers provide services;
how to guarantee that the service can be accessed by authorized entities. In the
literature, there have been a
large number of works proposed to answer these questions. We briefly introduce them
from the
following three perspectives.
5.2.1 Authentication and Trust. When assigning an IoT device to an edge server,
authenticating the identity of the owner to a specific edge server needs to be
considered [122]. To reduce
the authentication burden on IoT devices, Reference [135] designs an
electroencephalography authentication system by performing machine learning
algorithms on an edge server (i.e., laptop).
In this system, IoT devices capture the raw electroencephalography signals and
sends them to
a smart phone via Bluetooth. On a smart phone, the system interface application
receives electroencephalography signals and sends them to a laptop. The laptop uses
machine learning algorithms to make identification/authentication. To realize
efficient authentication, the authors of
Reference [71] develop a secure and efficient approach for fog computing, which
allows fog users
to mutually authenticate with each other under the authority of a cloud service
provider. The
proposed approach does not require a fog user to be included in a public key
infrastructure. The
fog user only needs to store one master secret key in the registration phase once
to mutually
authenticate with any other fog server. Moreover, the authors of Reference [51]
propose an efficient and responsive mutual authentication scheme by using elliptic
curve cryptography on edge
servers to protect smart grid system. Considering the scenario without a trust
third-party in an IoT
network, Reference [53] establishes trusted identities in disconnected edge
environments by combining identity-based cryptography with secure key generation
and exchange mechanisms. To
integrate different authentication protocols on heterogeneous IoT devices,
Reference [128] proposes a common identity and authentication scheme, which is able
to resist many typical attacks
in IoT scenarios, e.g., replay, man-in-the-middle, and masquerade attacks.
5.2.2 Privacy Preservation. As the edge devices are widely deployed to process the
generated
data from IoT devices (smart grid meters in home area networks, body-sensors that
collect health
information) and provide network services (such as data storage and data sharing),
a privacy leakage problem may arise during data processing provided at the network
edge. As a classical technique, homomorphic encryption has been applied to
providing privacy-preserving aggregation on
encrypted data at the network edge [161]. In Reference [148], the authors propose
an anonymous
and secure aggregation scheme by which edge servers can help IoT devices upload
their sensed
data to a service provider. The proposed scheme can well preserve the identities of
devices with
pseudonyms and guarantee the data secrecy based on homomorphic encryption,
simultaneously.
Lu et al. [92] present a privacy-preserving and efficient aggregation scheme for
communication in
smart grid. It structures multi-dimensional data by utilizing a super-increasing
sequence and encrypts the structured data with homomorphic cryptography techniques.
A homomorphic function
uses the encrypted data of the smart meters as input and generates the aggregated
and encrypted
results.
125:24
J. Ren et al.
Reference [91] proposes a privacy-preserving data aggregation scheme to aggregate

hybrid IoT
data and filter injected false data at the network edge. A fog-based de-duplicated
spatial crowdsourcing framework is designed to allocate task and sense data
deduplication security for fogassisted crowdsourcing system [106]. To realize
secure duplication in crowdsourcing environments
for storage, the authors of Reference [79] propose a hybrid secure deduplication
client-server protocol for untrustworthy fog storage scenarios. The authors of
Reference [19] propose a certificateless aggregate signcryption scheme for
preserving privacy in a vehicular crowdsensing system. In
addition, Reference [100] proposes an efficient data sharing scheme to securely
share data among
edge devices in a cloud-assisted IoT system.
5.2.3 Access Control. As an important research topic in IoT systems, access control
has
also attracted increasing attention under the hierarchical architecture of emerging
computing
paradigms [140]. In the area of e-healthcare systems, Reference [13] integrates the
access security broker to fog nodes and proposes an attribute-based access control
approach to realize fast
response for latency-sensitive e-healthcare applications and to prevent
unauthorized access to
health information. For providing secure services for heterogeneous resources, a
policy-based access control mechanism is developed to achieve secure collaboration
and interoperability in a distributed manner [52]. A holistic solution is exploited
to achieve multi-factor access control for
edge servers [56]. This solution enables layered security for both of the extrinsic
parties and the
storage service providers and is robust in resisting the colluding attacks by a
(small) fraction of
service providers.
Since IoT devices (such as wearable devices and wireless sensors) are usually
resource limited,
they lack powerful computation ability and face the security challenges to protect
themselves or
provide complex cryptographic algorithms to preserve their generated data.
Researchers integrate
edge computing to provide complementary security solutions for IoT devices to
reduce the computation burden. To decrease the computational cost of decryption,
Reference [182] introduces
the attribute-based encryption with outsourced decryption on edge servers to
offload the large
amount of computation. Reference [101] designs a privacy-preserving system to
enable the cloud
users to enhance the privacy and security of sensitive personal data. By combining
attribute-based
and role-based access control models, a new approach is proposed to preserve the
data privacy and
satisfy both of mandatory and discretionary access control requirements.
6
FUTURE RESEARCH DIRECTIONS
By fully exploiting the resources of end devices and the nearby

devices/infrastructures, the emerging network computing paradigms are paving the
road toward a more scalable and intelligent IoT
era. Although existing studies provide comprehensive investigations on different
network computing paradigms from various aspects, end-edge-cloud computing is still
in its infancy and needs
continuous research efforts to fuel the explosion of more intelligent IoT
applications. In this section, we briefly outline some future research directions
to foster further research.
6.1
Lightweight Virtualization for Network Computing
The dramatically increasing IoT devices have brought new challenges to fully
exploit the heterogeneous resources associated to them. The lightweight hardware
and high heterogeneity of IoT devices also bring various customized operating
systems and software, hindering the flexibility of IoT
services and service sharing among different IoT applications. Powered by the
advanced network
computing technologies, device virtualization solutions have shown the potentials
for providing
desired tradeoff between flexibility and performance. Redesigning the
virtualization methods for
lightweight IoT and edge devices becomes a growing trend in the research field of
virtualization.
125:25
Some emerging technologies, such as Docker and LXC containers, have been already
successfully
deployed on single-board computers to provide virtualized instances with an
efficient overhead.
As a representative of operating-system-level virtualization technologies,
containerization has
refashioned the world of software development by introducing flexibility and new
way of managing and distributing software. In Reference [126], the authors analyze
the relation between
the lightweight virtualization and edge computing. They have also answered how to
effectively
combine these two technologies and pointed out some classical application
scenarios. Morabito
et al. [103] propose two container-based IoT service provisioning frameworks and
compare their
performance based on a real IoT testbed. The one is for the case where two
cooperating devices
interact directly, and the other is for the case where a manager can supervise the
operations among
cooperating devices. Although existing works provide some valuable preliminary
results on applying container-based virtualization solutions in IoT scenarios,
there are still many technical challenges in developing more reliable and robust
lightweight virtualization technologies. For example,
how to orchestrate the edge resources and elements, how to monitor the performance
and status
of different service instances, how to provide security and privacy guarantees for
virtualized applications, as well as how to achieve efficient service migration
among different containers. These
remaining challenges leave a huge space for researchers to study more mature and
advanced lightweight virtualization technologies for network computing systems.
6.2 Block-streaming Service Loading
Besides running services on edge servers by lightweight virtualization (like
Software-as-a-Service
in cloud computing), dynamically loading on-demand services and executing them on
IoT devices
is an alternative solution for dynamic service provisioning in the IoT era. In
2003, the authors of
Reference [118] propose a stream way to execute the software on devices. In such a
way, the software can be executed when it is being streamed without waiting for
completing the whole process
of downloading, decompression, installation, and reconfiguration. A streaming
approach can reduce the experienced application loading time of users, since the
application can start running once
the first executable unit has been loaded into the memory. Recently, He et al. [67]
propose a new
cloud service model based on transparent computing, named Block-stream as a service
(BaaS),
for IoT devices. The BaaS model provides a framework for lightweight IoT devices to
load and
execute part of service codes on demand, by fully exploiting the advantages of
transparent computing in terms of cross-platform and dynamic service provisioning.
Inspired by the BaaS model,
Peng et al. [115] propose a transparent computing-based block-streaming application
execution
scheme, called BOAT, and implement it on a lightweight wearable platform. It can
remotely load a
part of necessary application codes from edge servers and execute the codes on IoT
devices locally.
Their experimental results demonstrate that BOAT can achieve significant
improvements in terms
of service loading delay and energy consumption when compared to the traditional
Softwareas-a-Service approach. Some similar concepts, like microservices [143],
have also been proposed
to improve the modularity, distribution, elasticity, and robustness of service
provisioning for IoT
devices by leveraging the capability of network computing.
Although block-streaming service loading has shown the potential to revolutionize
the service
provisioning way for lightweight IoT devices, continuous efforts should be devoted
in this research
direction to address some open challenges. The bandwidth-limited and unstable
wireless communications may make dynamic service loading intermittent and
consequently lead to poor quality
of experience (QoE). How to design efficient service prefetching mechanisms becomes
an important research direction to improve the QoE of users under the stochastic
wireless communication
environment. Moreover, since the whole service may be divided into different parts
to response
the service request of IoT devices, the integrity check and verification of the
service should be
125:26
J. Ren et al.
Fig. 7. APIs for end-edge-cloud orchestrated computing systems.
carefully redesigned to address the security risk. In addition, if the software of

IoT services is programmed in a modular way, then the efficiency of block-streaming
service loading can be further
enhanced. It means that standards or rules can be made to regulate the IoT software
programs to
guarantee the modularity of IoT services.
6.3 API Standards and Development for End-Edge-Cloud Orchestrated Computing
To standardize and facilitate application development for end-edge-cloud
orchestrated computing
systems, investigating the Application Programming Interface (API) standards and
developing flexible APIs have been regarded as a very important research direction.
A well-defined API consists
of subroutine definitions, communication protocols, and tools for building software
connections,
which can provide flexibility and seamless connectivity for different kinds of
devices and authorize other pieces of software to control the functionality of an
application or a service. Figure 7
illustrates how APIs work for the orchestration of end-edge-cloud systems [6]. As
shown in the
figure, the APIs developed for edge servers and the cloud should be carefully
designed and coordinated by considering their different resources and capabilities.
However, due to the complexity and
dynamics caused by the orchestration of end-edge-cloud resources, the design,
management, and
integration of APIs are facing great challenges. The first one is, from the
communication perspective, how to provide flexible and seamless connectivity for
highly mobile users to distributed edge
servers and edge servers to the cloud under the hybrid accessing technologies. The
second one is,
from the service perspective, how to design suitable APIs and
management/integration techniques
that can fully exploit and coordinated the differentiated resources and
capabilities of devices for
managing service provisioning and application development. Especially when
microservices are
becoming the main trend in the IoT era, dividing applications into feature
components and microservices drives API management and integration more significant
than ever. Components and
services that may be only parts of a single application need to be well integrated
for working in
a coordinated way and to deliver the requested capabilities for users. Thus, more
research efforts
should be devoted to the standardization and development of APIs for end-edge-cloud
orchestrated
computing.
125:27
6.4 Flexible Networked Operating System Design for Heterogeneous IoT Devices
The types of IoT devices range from ultra lightweight RFID tags to some powerful
smart phones
and vehicles. The hardware heterogeneity of IoT devices also leads to a wide
variety of IoT operating systems in the community. For most of lightweight IoT
devices, embedded realtime OSes,
such as TinyOS and Contiki, are adopted to provide specific functionalities with
strict delay constraints. But for the powerful IoT devices, general OSes, such as
Android and iOS, are widely
applied to support resource sharing and multi-task processing. Diverse OSes cause
an inevitable
limitation that IoT services cannot be shared among different IoT devices. And in
most cases, even
with the same operating system kernel, the OSes developed by different vendors
cannot support
cross-platform service sharing. Therefore, designing a flexible operating system
that can be well
applied on heterogeneous IoT devices attracts increasing attention from both of the
academia and
industry. For example, Google is developing a new IoT OS, named Fuchsia, which is
based on a new
kernel named Magenta. It aims to provide flexible deployment on heterogeneous IoT
devices by
using a micro-kernel and a large set of incremental services and drivers. IoT
devices can choose to
add functionalities on the micro-kernel according to their capabilities and
requirements, such that
cross-platform services can be well supported by any kind of devices. However, with
the development of network computing technologies, the network-connected IoT
devices can be regarded as
an integrated system. We can also leverage the power of network computing to change
the traditional way of OS design. Flexible networked OS, where OS kernels or
functionalities are distributed
over different connected devices, may be a new trend for IoT operating system
design.
6.5
Lightweight and Distributed Machine Learning for Realtime Intelligence
Recent advances in machine learning, especially in deep learning, have surged the
development
of intelligent IoT applications, such as mobile face recognition, video analytics,
and autonomous
path planning drones. These kinds of intelligent applications rely on
computationally intensive
machine learning algorithms and require real-time processing. Due to the hardware
limitations,
however, mobile and IoT devices are facing great challenges to fulfill stringent
QoS requirements
of intelligent applications. For example, video analytics, which usually involves
complex deep neural networks, far exceeds the computing capabilities of the end
devices. Meanwhile, offloading the
videos to the cloud server incurs unbearable burdens on the core network and
produces unacceptable delay, which consequently hinders the efficiency of video
analytics. Leveraging network
computing to directly perform machine learning and data analysis at mobile
terminals or the network edge becomes a promising way to provide computation
augmenting and timely intelligent
services for mobile and IoT devices. This has also been regarded as the “killer
application” of edge
computing.
However, most IoT devices and edge servers are not capable enough to support the
machine
learning and data analytical algorithms conventionally deployed in the resource-
extensive cloud.
To well support intelligent mobile and IoT applications, therefore, there is a
significant need for
exploring the implementation of lightweight and distributed machine learning models
at IoT devices and the network edge, studying efficient resource allocation
algorithms for real-time machine learning among devices, edge servers, and the
cloud, as well as designing collaborative and
distributed data analyzing architectures for network computing paradigms. Recently,
Google proposed a decentralized machine learning approach, named federated
learning, to train the data distributed among lightweight IoT devices then to merge
the aggregated locally computed updates to
one shared model [78]. It introduces a federated learning algorithm that combines
local stochastic
gradient descent training on each client within relatively few communication rounds
where the
central cloud server performs model averaging. Following the idea of distributed
learning, Hardy
125:28
J. Ren et al.
et al. [45] propose AdaComp, a distributed deep learning algorithm running on edge
devices, which
can compress the edge updates to the model on cloud servers. These existing works
have pointed
out a potential research direction for implementing lightweight and distributed
machine learning
algorithms, with the aim of achieving efficient and timely data training and
analysis of IoT devices
and edge servers, and reducing the communication rounds between the edge side and
the cloud
side. In such a way, the privacy of personal data can also be well preserved, since
the data traffic
originally transmitted to the cloud for model training and analysis can be
significantly reduced.
6.6 Advanced Security and Privacy Preservation Techniques for Network Computing
Since lightweight IoT devices are generally limited by computing capability,
storage, and energy
supply, traditional security solutions, e.g., complex cryptographic solutions, may
not be applicable
for IoT devices. Thus, how to leverage network computing to offload the security
protection burden
of IoT platforms becomes a potential and promising research direction. As we
discussed in the
previous section, there have been many existing works focused on offloading the
authentication
and encryption burden from lightweight IoT devices to edge servers. These solutions
provide some
valuable insights to address the security and privacy problems in current IoT
applications.
However, compared to the resourceful cloud servers, the resource-constrained edge
servers are
still vulnerable to different kinds of attacks. It consequently is necessary that
future security techniques should fully leverage the capabilities of edge servers
to provide enhanced security protection for IoT devices and also can guarantee the
security of the edge servers. Thus, how to efficiently
utilize the three-tire framework, i.e., end-edge-cloud, to collaboratively achieve
advanced security
and privacy preservation for the whole IoT platform becomes an important research
direction for
future studies. Moreover, for ultra-light-weight IoT devices, e.g., RFID tags, even
very lightweight
security approaches cannot be efficiently implemented. Since such kinds of IoT
devices will take
a significant portion of “connected things” and face great security and privacy
leakage risks in
the IoT era, more attention should be paid to leveraging network computing for
addressing this
challenge.
7
CONCLUSION
In this survey, we have provided an overview and comparison of the emerging network
computing
paradigms, including transparent computing, mobile edge computing, fog computing,
and cloudlet
computing. The common feature of the computing paradigms is to orchestrate the
computing
and storage resources of end devices, edge servers, and the cloud to improve the
performance
of hierarchical IoT systems. From this perspective, we have presented a
comprehensive survey on
some hot research issues of the computing paradigms, in terms of computation
offloading, caching,
as well as security and privacy. Finally, we have discussed some potential future
directions for
emerging and evolving research.
REFERENCES
[1] Mobile Edge Computing Introductory Technical White Paper. 2014. Retrieved from
https://portal.etsi.org/portals/
0/tbpages/mec/docs/mobile-edge_computing_-
_introductory_technical_white_paper_v1%2018-09-14.pdf.
[2] Mobile-Edge Computing (MEC); Service Scenarios. 2015. Retrieved from
http://www.etsi.org/deliver/etsi_gs/
MEC-IEG/001_099/004/01.01.01_60/gs_MEC-IEG004v010101p.pdf.
[3] Edge computing prepares for a multi-access future. 2016. Retrieved from
http://www.telecomtv.com/articles/mec/
edge-computing-prepares-for-a-multi-access-future-13986/.
[4] Mobile-Edge Computing (MEC); Technical Requirements. 2016. Retrieved from
http://www.etsi.org/deliver/etsi_gs/
MEC/001_099/002/01.01.01_60/gs_MEC002v010101p.pdf.
[5] Mobile-Edge Computing (MEC); Terminology. 2016. Retrieved from
http://www.etsi.org/deliver/etsi_gs/MEC/001_
099/001/01.01.01_60/gs_MEC001v010101p.pdf.
125:29
[6] APIs for IOT and FOG computing. 2017. Retrieve from
https://www.vanrish.com/blog/2017/12/17/apis-for-iotand-fog-computing/.
[7] Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update,
2016-2021. 2017. Retrieved from https://
www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-
index-vni/mobile-white-paperc11-520862.html.
[8] Cloud Adoption Is Growing But Forecasts Differ on How Much. 2017. Retrieved
from http://fortune.com/2017/02/
22/cloud-growth-forecast-gartner/.
[9] Mobile Edge Computing; Market Acceleration; MEC Metrics Best Practice and
Guidelines. 2017. Retrieved from
http://www.etsi.org/deliver/etsi_gs/MEC-IEG/001_099/006/01.01.01_60/gs_MEC-
IEG006v010101p.pdf.
[10] OpenFog Reference Architecture for Fog Computing. 2017. Retrieved from
https://www.openfogconsortium.org/
wp-content/uploads/OpenFog_Reference_Architecture_2_09_17-FINAL.pdf.
[11] Hasti Ahlehagh and Sujit Dey. 2014. Video-aware scheduling and caching in the
radio access network. IEEE/ACM
Trans. Netw. 22, 5 (2014), 1444–1462.
[12] Arif Ahmed and Ejaz Ahmed. 2016. A survey on mobile edge computing. In
Proceedings of the 2016 10th International
Conference on Intelligent Systems and Control (ISCO’16). 1–8.
DOI:https://doi.org/10.1109/ISCO.2016.7727082
[13] Mohammed Alshiky Aisha, M. Buhari Seyed, and Barnawi Ahmed. 2017. Attribute-
based access control (ABAC) for
EHR in fog computing environment. Int. J. Cloud Comput.: Serv. Arch. 7, 1 (2017),
9–16.
[14] Moustafa Ammar, Mohamed Rizk, Ayman Abdel-Hamid, and Ahmed K. Aboul-Seoud.
2016. A framework for security enhancement in SDN-based datacenters. In Proceedings
of the IFIP International Conference on New Technologies,
Mobility and Security (NTMS’16). IEEE, 1–4.
[15] Arash Asadi, Qing Wang, and Vincenzo Mancuso. 2014. A survey on device-to-
device communication in cellular
networks. IEEE Commun. Surv. Tutor. 16, 4 (2014), 1801–1819.
[16] Bahar Azari, Osvaldo Simeone, Umberto Spagnolini, and Antonia M. Tulino. 2016.
Hypergraph-based analysis of
clustered co-operative beamforming with application to edge caching. IEEE Wireless
Commun. Lett. 5, 1 (2016), 84–87.
[17] Bo Bai, Li Wang, Zhu Han, Wei Chen, and Tommy Svensson. 2016. Caching based
socially-aware D2D communications in wireless content delivery networks: a
hypergraph framework. IEEE Wireless Commun. 23, 4 (2016), 74–81.
[18] Ejder Bastug, Mehdi Bennis, and Mérouane Debbah. 2014. Living on the edge: The
role of proactive caching in 5G
wireless networks. IEEE Commun. Mag. 52, 8 (2014), 82–89.
[19] Sultan Basudan, Xiaodong Lin, and Karthik Sankaranarayanan. 2017. A privacy-
preserving vehicular crowdsensing
based road surface condition monitoring system using fog computing. IEEE IoT J. 4,
3 (2017), 772–782.
[20] Suzhi Bi and Ying Jun Zhang. 2018. Computation rate maximization for wireless
powered mobile-edge computing
with binary computation offloading. IEEE Trans. Wireless Commun. 17, 6 (Jun. 2018),
4177–4190. DOI:https://doi.org/
10.1109/TWC.2018.2821664
[21] Flavio Bonomi, Rodolfo Milito, Preethi Natarajan, and Jiang Zhu. 2014. Fog
Computing: A Platform for Internet of Things and Analytics. Springer International
Publishing, Cham, 169–186. DOI:https://doi.org/10.1007/
978-3-319-05029-4_7
[22] Flavio Bonomi, Rodolfo Milito, Jiang Zhu, and Sateesh Addepalli. 2012. Fog
computing and its role in the Internet
of Things. In Proceedings of the Workshop on Mobile Cloud Computing (MCC’12). ACM,
New York, NY, 13–16.
[23] Sem Borst, Varun Gupta, and Anwar Walid. 2010. Distributed caching algorithms
for content distribution networks.
In Proceedings of the IEEE International Conference on Computer Communications
(INFOCOM’10). IEEE, 1–9.
[24] Piotr Borylo, Artur Lason, Jacek Rzasa, Andrzej Szymanski, and Andrzej
Jajszczyk. 2016. Energy-aware fog and cloud
interplay supported by wide area software defined networking. In Proceedings of the
IEEE International Conference
on Communications (ICC’16). IEEE, 1–7.
[25] A. Bourdena, C. X. Mavromoustakis, G. Mastorakis, J. J. P. C. Rodrigues, and
C. Dobre. 2015. Using socio-spatial
context in mobile cloud process offloading for energy conservation in wireless
devices. IEEE Trans. Cloud Comput.
1 (2015).
[26] Shiwei Cao, Xiaofeng Tao, Yanzhao Hou, and Qimei Cui. 2015. An energy-optimal
offloading algorithm of mobile
computing based on HetNets. In Proceedings of the IEEE International Conference on
Connected Vehicles and Expo
(ICCVE’15). IEEE, 254–258.
[27] Sami Yangui Roch H. Glitho Monique J. Morrow Carla Mouradian, Diala Naboulsi
and Paul A. Polakos. 2018. A
comprehensive survey on fog computing: State-of-the-art and research challenges.
IEEE Commun. Surv. Tutor. 20, 1
(2018), 416–464.
[28] Meeyoung Cha, Haewoon Kwak, Pablo Rodriguez, Yong-Yeol Ahn, and Sue Moon.
2009. Analyzing the video popularity characteristics of large-scale user generated
content systems. IEEE/ACM Trans. Netw. 17, 5 (2009), 1357–1370.
[29] Omar Chakroun and Soumaya Cherkaoui. 2016. Resource allocation for delay
sensitive applications in mobile cloud
computing. In Proceedings of the IEEE Conference on Local Computer Networks
(LCN’16). IEEE, 615–618.
[30] Binqiang Chen, Chenyang Yang, and Gang Wang. 2016. Cooperative device-to-
device communications with caching.
In Proceedings of the IEEE Vehicular Technology Conference (VTC’16). IEEE, 1–5.
125:30
J. Ren et al.
[31] Binqiang Chen, Chenyang Yang, and Gang Wang. 2017. High throughput
opportunistic cooperative device-to-device
communications with caching. IEEE Trans. Vehic. Technol. 66, 8 (2017), 7527–7539.
[32] Di Chen, Stephan Schedler, and Volker Kuehn. 2016. Backhaul traffic balancing
and dynamic content-centric clustering for the downlink of fog radio access
network. In Proceedings of the IEEE SPAWC. IEEE, 1–5.
[33] Guangyu Chen, B.-T. Kang, Mahmut Kandemir, Narayanan Vijaykrishnan, Mary Jane
Irwin, and Rajarathnam Chandramouli. 2004. Studying energy trade offs in offloading
computation/compilation in java-enabled mobile devices.
IEEE Trans. Parallel Distrib. Syst. 15, 9 (2004), 795–809.
[34] Lixing Chen, Sheng Zhou, and Jie Xu. 2018. Computation peer offloading for
energy-constrained mobile edge computing in small-cell networks. IEEE/ACM Trans.
Netw. 26, 4 (2018), 1619–1632. DOI:https://doi.org/10.1109/TNET.
2018.2841758
[35] Min Chen, Yixue Hao, Chin-Feng Lai, Di Wu, Yong Li, and Kai Hwang. 2018.
Opportunistic task scheduling over
co-located clouds in mobile environment. IEEE Trans. Serv. Comput. 11, 3 (2018),
549–561.
[36] M. Chen, Y. Hao, Y. Li, C. Lai, and D. Wu. 2015. On the computation offloading
at ad hoc cloudlet: Architecture and
service modes. IEEE Commun. Mag. 53, 6 (Jun. 2015), 18–24.
DOI:https://doi.org/10.1109/MCOM.2015.7120041
[37] Mingzhe Chen, Walid Saad, Changchuan Yin, and Mérouane Debbah. 2017. Echo
state networks for proactive
caching in cloud-based radio access networks with mobile users. IEEE Trans.
Wireless Commun. 16, 6 (2017), 3520–
3535.
[38] Mingkai Chen, Lei Wang, and Jianxin Chen. 2016. Users-media cloud assisted D2D
communications for distributed
caching underlaying cellular network. Chin. Commun. 13, 8 (2016), 13–23.
[39] Meng-Hsi Chen, Min Dong, and Ben Liang. 2016. Joint offloading decision and
resource allocation for mobile cloud
with computing access point. In Proceedings of the IEEE International Conference on
Acoustics, Speech, and Signal
Processing (ICASSP’16). IEEE, 3516–3520.
[40] Meng-Hsi Chen, Ben Liang, and Min Dong. 2015. A semidefinite relaxation
approach to mobile cloud offloading with
computing access point. In Proceedings of the IEEE International Workshop on Signal
Processing Advances in Wireless
Communications (SPAWC’15). IEEE, 186–190.
[41] Xu Chen, Lei Jiao, Wenzhong Li, and Xiaoming Fu. 2016. Efficient multi-user
computation offloading for mobile-edge
cloud computing. IEEE/ACM Trans. Netw. 24, 5 (2016), 2795–2808.
[42] Mung Chiang, Sangtae Ha, I. Chih-Lin, Fulvio Risso, and Tao Zhang. 2017.
Clarifying fog computing and networking:
10 questions and answers. IEEE Commun. Mag. 55, 4 (2017), 18–20.
[43] Byung-Gon Chun, Sunghwan Ihm, Petros Maniatis, Mayur Naik, and Ashwin Patti.
2011. Clonecloud: Elastic execution between mobile device and cloud. In Proceedings
of the the 6th Conference on Computer Systems. ACM, 301–314.
[44] Byung-Gon Chun and Petros Maniatis. 2010. Dynamically partitioning
applications between weak devices and
clouds. In Proceedings of the 1st ACM Workshop on Mobile Cloud Computing &
Services: Social Networks and Beyond.
ACM, 7.
[45] Hardy Corentin, Le Merrer Erwan, and Sericola Bruno. 2017. Distributed deep
learning on edge-devices: Feasibility
via adaptive compression. In Proceedings of the IEEE International Symposium on
Network Computing and Applications
(NCA’17). IEEE, 1–8.
[46] Eduardo Cuervo, Aruna Balasubramanian, Dae-ki Cho, Alec Wolman, Stefan Saroiu,
Ranveer Chandra, and Paramvir
Bahl. 2010. MAUI: Making smartphones last longer with code offload. In Proceedings
of the ACM International Conference on Mobile Systems, Applications, and Services
(Mobisys’10). ACM, 49–62.
[47] Amir Vahid Dastjerdi and Rajkumar Buyya. 2016. Fog computing: Helping the
Internet of Things realize its potential.
Computer 49, 8 (2016), 112–116.
[48] Maofei Deng, Hui Tian, and Bo Fan. 2016. Fine-granularity based application
offloading policy in cloud-enhanced
small cell networks. In Proceedings of the IEEE International Conference on
Communications (ICC’16). IEEE, 638–643.
[49] Haichuan Ding and Yuguang Fang. 2018. Virtual infrastructure at traffic
lights: Vehicular temporary storage assisted
data transportation at signalized intersections. IEEE Trans. Vehic. Technol. 67, 12
(2018), 12452–12456.
[50] Haichuan Ding, Yuanxiong Guo, Xuanheng Li, and Yuguang Fang. 2018. Beef up the
edge: Spectrum-aware placement of edge computing services for the Internet of
Things. IEEE Trans. Mobile Comput. (2018). DOI:https://doi.org/
10.1109/TMC.2018.2883952
[51] Abebe Abeshu Diro, Naveen Chilamkurti, and Neeraj Kumar. 2017. Lightweight
cybersecurity schemes using elliptic
curve cryptography in publish-subscribe fog computing. Mobile Netw. Appl. 22, 5
(2017), 848–858.
[52] Clinton Dsouza, Gail-Joon Ahn, and Marthony Taguinod. 2014. Policy-driven
security management for fog computing: Preliminary framework and a case study. In
Proceedings of the IEEE International Conference on Information
Reuse and Information (IRI’14). IEEE, 16–23.
[53] Sebastián Echeverría, Dan Klinedinst, Keegan Williams, and Grace A. Lewis.
2016. Establishing trusted identities
in disconnected edge environments. In Proceedings of the IEEE/ACM Symposium on Edge
Computing (SEC’16). IEEE,
51–63.
125:31
[54] Khalid Elgazzar, Patrick Martin, and Hossam S. Hassanein. 2016. Cloud-assisted
computation offloading to support
mobile services. IEEE Trans. Cloud Comput. 4, 3 (2016), 279–292.
[55] Jeffrey Erman, Alexandre Gerber, Mohammad Hajiaghayi, Dan Pei, Subhabrata Sen,
and Oliver Spatscheck. 2011.
To cache or not to cache: The 3G case. IEEE Internet Comput. 15, 2 (2011), 27–34.
[56] Ertem Esiner and Anwitaman Datta. 2016. Layered security for storage at the
edge: On decentralized multifactor access control. In Proceedings of the ACM
International Conference on Distributed Computing and Networking
(ICDCN’16). ACM, 9.
[57] Jose Oscar Fajardo, Ianire Taboada, and Fidel Liberal. 2015. Radio-aware
service-level scheduling to minimize downlink traffic delay through mobile edge
computing. In Proceedings of the 7th EAI International Conference on Mobile
Networks and Management. Springer, 121–134.
[58] Yuan Gao, Yaoxue Zhang, and Yuezhi Zhou. 2012. A cache management strategy for
transparent computing storage
system. In Proceedings of the International Conference on New Trends in Computing
Sciences (ICTCS’12). Springer,
651–658.
[59] Ioana Giurgiu, Oriana Riva, Dejan Juric, Ivan Krivulev, and Gustavo Alonso.
2009. Calling the cloud: Enabling mobile
phones as interfaces to cloud applications. In Proceedings of the International
Middleware Conference (Middleware’09).
Springer-Verlag New York, 5.
[60] Negin Golrezaei, Parisa Mansourifard, Andreas F Molisch, and Alexandros G
Dimakis. 2014. Base-station assisted
device-to-device communications for high-throughput wireless video networks. IEEE
Trans. Wireless Commun. 13,
7 (2014), 3665–3676.
[61] Negin Golrezaei, Karthikeyan Shanmugam, Alexandros G. Dimakis, Andreas F.
Molisch, and Giuseppe Caire. 2012.
Femtocaching: Wireless video content delivery through distributed caching helpers.
In Proceedings of the IEEE International Conference on Computer Communications
(INFOCOM’12). IEEE, 1107–1115.
[62] Andre S. Gomes, Bruno Sousa, David Palma, Vitor Fonseca, Zhongliang Zhao,
Edmundo Monteiro, Torsten Braun,
Paulo Simoes, and Luis Cordeiro. 2017. Edge caching with mobility prediction in
virtualized LTE mobile networks.
Future Gener. Comput. Syst. 70, 5 (2017), 148–162.
[63] Mohammad Goudarzi, Zeinab Movahedi, and Masoud Nazari. 2016. Mobile cloud
computing: A multisite computation offloading. In Proceedings of the IEEE
International Conference on Imaging Systems and Techniques (IST’16). IEEE,
660–665.
[64] Jingxiong Gu, Wei Wang, Aiping Huang, Hangguan Shan, and Zhaoyang Zhang. 2014.
Distributed cache replacement for caching-enable base stations in cellular
networks. In Proceedings of the IEEE International Conference on
Communications (ICC’14). IEEE, 2648–2653.
[65] Hongzhi Guo and Jiajia Liu. 2018. Collaborative computation offloading for
multiaccess edge computing over fiber–
wireless networks. IEEE Trans. Vehic. Technol. 67, 5 (May 2018), 4514–4526.
[66] Kiryong Ha and Mahadev Satyanarayanan. 2015. OpenStack++ for Cloudlet
Deployment. Technical Report. http://
elijah.cs.cmu.edu/DOCS/CMU-CS-15-113.pdf
[67] J. He, Y. Zhang, J. Lu, M. Wu, and F. Huang. 2018. Block-stream as a service:
A more secure, nimble, and dynamically
balanced cloud service model for ambient computing. IEEE Netw. 32, 1 (2018), 126–
132.
[68] Dinh T. Hoang, Chonho Lee, Dusit Niyato, and Ping Wang. 2013. A survey of
mobile cloud computing: Architecture,
applications, and approaches. Wireless Commun. Mobile Comput. 13, 18 (2013), 1587–
1611.
[69] Farhoud Hosseinpour, Payam Vahdani Amoli, Juha Plosila, Timo Hämäläinen, and
Hannu Tenhunen. 2016. An intrusion detection system for fog computing and IoT based
logistic systems using a smart data approach. Int. J. Dig.
Content Technol. Appl. 10, 5 (2016), 34–46.
[70] Yunchao Hu, Milan Patel, Dario Sabella, Nurit Sprecher, and Valerie Young.
2015. ETSI. White Paper No. 11. Mobile
Edge Computing a Key Technology Towards 5G [S]. DOI:https://doi.org/10.1007/978-3-
319-05029-4_7
[71] Maged Ibrahim. 2016. Octopus: An edge-fog mutual authentication scheme. Int.
J. Netw. Secur. 18, 6 (2016), 1089–
1101.
[72] Andriana Ioannou and Stefan Weber. 2016. A survey of caching policies and
forwarding mechanisms in informationcentric networking. IEEE Commun. Surv. Tutor.
18, 4 (2016), 2847–2886.
[73] Mingyue Ji, Giuseppe Caire, and Andreas F. Molisch. 2016. Wireless device-to-
device caching networks: Basic principles and system performance. IEEE J. Select.
Areas Commun. 34, 1 (2016), 176–189.
[74] Shiwei Jia, Yuan Ai, Zhongyuan Zhao, Mugen Peng, and Chunjing Hu. 2016.
Hierarchical content caching in fog
radio access networks: Ergodic rate and transmit latency. Chin. Commun. 13, 12
(2016), 1–14.
[75] Wei Jiang, Gang Feng, and Shuang Qin. 2017. Optimal cooperative content
caching and delivery policy for heterogeneous cellular networks. IEEE Trans. Mobile
Comput. 16, 5 (2017), 1382–1393.
[76] Sladana Josĭlo and György Dán. 2019. Selfish decentralized computation
offloading for mobile cloud computing in
dense wireless networks. IEEE Trans. Mobile Comput. 18, 1 (2019), 207–220.
DOI:https://doi.org/10.1109/TMC.2018.
2829874
125:32
J. Ren et al.
[77] Sarang Kahvazadeh, Vitor B. Souza, Xavi Masip-Bruin, Eva Marn-Tordera, Jordi
Garcia, and Rodrigo Diaz. 2017.
Securing combined fog-to-cloud system through SDN approach. In Proceedings of the
Workshop on CrossCloud Infrastructures & Platforms (CrossCloud’17). ACM, 2.
[78] Jakub Koneáý, H. Brendan, McMahan, Daniel Ramage, and Peter Richtárik. 2016.
Federated optimization: Distributed
machine learning for on-device intelligence. arXiv:1610.02527v1 (10 2016).
[79] Dongyoung Koo, Youngjoo Shin, Joobeom Yun, and Junbeom Hur. 2016. A hybrid
deduplication for secure and
efficient data outsourcing in fog computing. In Proceedings of the IEEE
International Conference on Cloud Computing
Technology and Science (CloudCom’16). IEEE, 285–293.
[80] Sokol Kosta, Andrius Aucinas, Pan Hui, Richard Mortier, and Xinwen Zhang.
2012. Thinkair: Dynamic resource
allocation and parallel execution in the cloud for mobile code offloading. In
Proceedings of the IEEE International
Conference on Computer Communications (INFOCOM’12). IEEE, 945–953.
[81] Wenyuan Kuang, Yaoxue Zhang, Yuezhi Zhou, and Huajie Yang. 2007. RBIS:
Security enhancement for MRBP and
MRBP2 using integrity check. J. Chin. Comput. Syst. 28, 2 (2007), 251–254.
[82] D. Van Le and C. Tham. 2018. A deep reinforcement learning based offloading
scheme in ad-hoc mobile clouds. In
Proceedings of the IEEE International Conference on Computer Communications
(INFOCOM’18). 760–765.
[83] Hongjia Li and Dan Hu. 2016. Mobility prediction based seamless RAN-cache
handover in HetNet. In Proc. IEEE
Wireless Communications and Networking Conference (WCNC’16). IEEE, 1–7.
[84] Xue Lin, Yanzhi Wang, Qing Xie, and Massoud Pedram. 2015. Task scheduling with
dynamic voltage and frequency
scaling for energy minimization in the mobile cloud computing environment. IEEE
Trans. Serv. Comput. 8, 2 (2015),
175–186.
[85] Ying-Dar Lin, Edward T-H Chu, Yuan-Cheng Lai, and Ting-Jun Huang. 2015. Time-
and-energy-aware computation
offloading in handheld devices to coprocessors and clouds. IEEE Syst. J. 9, 2
(2015), 393–405.
[86] Juan Liu, Yuyi Mao, Jun Zhang, and Khaled B. Letaief. 2016. Delay-optimal
computation task scheduling for mobileedge computing systems. In Proceedings of the
IEEE International Symposium on Information Theory (ISIT’16). IEEE,
1451–1455.
[87] Jinzhao Liu, Yuezhi Zhou, and Di Zhang. 2016. TranSim: A simulation framework
for cache-enabled transparent
computing systems. IEEE Trans. Comput. 65, 10 (2016), 3171–3183.
[88] Liqing Liu, Zheng Chang, and Xijuan Guo. 2018. Socially aware dynamic
computation offloading scheme for fog
computing system with energy harvesting devices. IEEE IoT J. 5, 3 (Jun. 2018),
1869–1879.
[89] Yanchen Liu, Myung J Lee, and Yanyan Zheng. 2016. Adaptive multi-resource
allocation for cloudlet-based mobile
cloud computing system. IEEE Trans. Mobile Comput. 15, 10 (2016), 2398–2410.
[90] Garcia Pedro Lopez, Alberto Montresor, Dick Epema, Anwitaman Datta, Teruo
Higashino, Adriana Iamnitchi, Marinho Barcellos, Pascal Felber, and Etienne
Riviere. 2015. Edge-centric computing: Vision and challenges. SIGCOMM
Comput. Commun. Rev. 45, 5 (2015), 37–42.
[91] Rongxing Lu, Kevin Heung, Arash Habibi Lashkari, and Ali A. Ghorbani. 2017. A
lightweight privacy-preserving
data aggregation scheme for fog computing-enhanced IoT. IEEE Access 5 (2017), 3302–
3312.
[92] Rongxing Lu, Xiaohui Liang, Xu Li, Xiaodong Lin, and Xuemin Shen. 2012. Eppa:
An efficient and privacy-preserving
aggregation scheme for secure smart grid communications. IEEE Trans. Parallel
Distrib. Syst. 23, 9 (2012), 1621–1631.
[93] Pavel Mach and Zdenek Becvar. 2016. Cloud-aware power control for real-time
application offloading in mobile
edge computing. Trans. Emerg. Telecommun. Technol. 27, 5 (2016), 648–661.
[94] S. Eman Mahmoodi, R. N. Uma, and K. P Subbalakshmi. 2019. Optimal joint
scheduling and cloud offloading for
mobile applications. IEEE Trans. Cloud Comput. 7, 2 (2019), 301–313.
[95] Yuyi Mao, Changsheng You, Jun Zhang, Kaibin Huang, and Khaled B. Letaief.
2017. Mobile edge computing: Survey
and research outlook. arXiv preprint arXiv:1701.01090 (2017).
[96] Yuyi Mao, Changsheng You, Jun Zhang, Kaibin Huang, and Khaled B. Letaief.
2017. A survey on mobile edge computing: The communication perspective. IEEE
Commun. Surv. Tutor. 19, 4 (2017), 2322–2358.
[97] Yuyi Mao, Jun Zhang, and Khaled B. Letaief. 2016. Dynamic computation
offloading for mobile-edge computing
with energy harvesting devices. IEEE J. Sel. Areas Commun. 34, 12 (2016), 3590–
3605.
[98] Yuyi Mao, Jun Zhang, S. H. Song, and Khaled B. Letaief. 2016. Power-delay
tradeoff in multi-user mobile-edge computing systems. arXiv preprint
arXiv:1609.06027 (2016).
[99] Yuyi Mao, Jun Zhang, S. H. Song, and Khaled B. Letaief. 2017. Stochastic joint
radio and computational resource
management for multi-user mobile-edge computing systems. IEEE Trans. Wireless
Commun. 16, 9 (2017), 5994–6009.
[100] Muhammad Baqer Mollah, Md Abul Kalam Azad, and Athanasios Vasilakos. 2017.
Secure data sharing and searching
at the edge of cloud-assisted Internet of Things. IEEE Cloud Comput. 4, 1 (2017),
34–42.
[101] Ei Ei Mon and Thinn Thu Naing. 2011. The privacy-aware access control system
using attribute-and role-based access control in private cloud. In Proceedings of
the IEEE International Conference on Broadband Network & Multimedia
Technology (IC-BNMT’11). IEEE, 447–451.
125:33
[102] Diego Montero, Marcelo Yannuzzi, Adrian Shaw, Ludovic Jacquin, Antonio
Pastor, Rene Serral-Gracia, Antonio
Lioy, Fulvio Risso, Cataldo Basile, Roberto Sassu, et al. 2015. Virtualized
security at the network edge: A user-centric
approach. IEEE Commun. Mag. 53, 4 (2015), 176–186.
[103] R. Morabito, I. Farris, A. Iera, and T. Taleb. 2017. Evaluating performance
of containerized IoT services for clustered
devices at the network edge. IEEE IoT J. 4, 4 (2017), 1019–1030.
[104] Chiang Mung, Ha Sangtae, I Chih-lin, Risso Fulvio, and Zhang Tao. 2017.
Clarifying fog computing and networking:
10 questions and answers. IEEE Commun. Mag. 55, 4 (Apr. 2017), 18–20.
DOI:https://doi.org/10.1109/MCOM.2017.
7901470
[105] Olga Muñoz, Antonio Pascual Iserte, Josep Vidal, and Marc Molina. 2014.
Energy-latency trade-off for multiuser
wireless computation offloading. In Proceedings of the IEEE Wireless Communications
and Networking Conference
Workshop (WCNCW’14). IEEE, 29–33.
[106] Jianbing Ni, Xiaodong Lin, Kuan Zhang, and Yong Yu. 2016. Secure and
deduplicated spatial crowdsourcing: A fogbased approach. In Proceedings of the IEEE
Global Communications Conference (GLOBECOM’16). 1–6.
[107] Ruifang Niu, Wenfang Song, and Yong Liu. 2013. An energy-efficient multisite
offloading algorithm for mobile devices. Int. J. Distrib. Sens. Netw. 9, 3 (2013),
1–6.
[108] OECI. 2017. Retrieved from http://openedgecomputing.org/index.html.
[109] Opeyemi Osanaiye, Shuo Chen, Zheng Yan, Rongxing Lu, Kim Choo, and Mqhele
Dlodlo. 2017. From cloud to fog
computing: A review and a conceptual live VM migration framework. IEEE Access
(2017).
[110] Jesus Pacheco and Salim Hariri. 2016. IoT security framework for smart cyber
infrastructures. In Proceedings
of the IEEE International Workshops on Foundations and Applications of Self*
Systems (FAS*W’16). IEEE, 242–
247.
[111] Seok-Hwan Park, Osvaldo Simeone, and Shlomo Shamai. 2016. Joint cloud and
edge processing for latency minimization in fog radio access networks. In
Proceedings of the IEEE Workshop on Signal Processing Advances in Wireless
Communications (SPAWC’16). IEEE, 1–5.
[112] Georgios Paschos, Ejder Bastug, Ingmar Land, Giuseppe Caire, and Mérouane
Debbah. 2016. Wireless caching: Technical misconceptions and business barriers.
IEEE Commun. Mag. 54, 8 (2016), 16–22.
[113] Mach Pavel and Becvar Zdenek. 2017. Mobile edge computing: A survey on
architecture and computation offloading.
IEEE Commun. Surv. Tutor. 19, 3 (Thirdquarter 2017), 1628–1656.
[114] Mugen Peng, Shi Yan, Kecheng Zhang, and Chonggang Wang. 2016. Fog-computing-
based radio access networks:
Issues and challenges. IEEE Netw. 30, 4 (2016), 46–53.
[115] Xuhong Peng, Ju Ren, Liang She, Deyu Zhang, Jie Li, and Yaoxue Zhang. 2018.
BOAT: A block-streaming app execution scheme for lightweight IoT devices. IEEE IoT
J. 5, 3 (2018), 1816–1829.
[116] Konstantinos Poularakis, George Iosifidis, Vasilis Sourlas, and Leandros
Tassiulas. 2014. Multicast-aware caching
for small cell networks. In Proceedings of the IEEE Wireless Communications and
Networking Conference (WCNC’14).
IEEE, 2300–2305.
[117] Konstantinos Poularakis, George Iosifidis, and Leandros Tassiulas. 2013.
Approximation caching and routing
algorithms for massive mobile data delivery. In Proceedings of the IEEE Global
Communications Conference
(GLOBECOM’13). IEEE, 3534–3539.
[118] Kuacharoen Pramote, J. Mooney Vincent, and K. Madisetti Vijay. 2003.
Identity-based authentication scheme for the
Internet of Things. In Proceedings of the Conference on Design, Automation and Test
in Europe Conference (DATE’03).
IEEE, 912–917.
[119] Lingjun Pu, Xu Chen, Jingdong Xu, and Xiaoming Fu. 2016. D2D fogging: An
energy-efficient and incentive-aware
task offloading framework via network-assisted D2D collaboration. IEEE J. Sel.
Areas Commun. 34, 12 (2016), 3887–
3901.
[120] Praveen Kumar Rajendran. 2015. Hybrid intrusion detection algorithm for
private cloud. Ind. J. Sci. Technol. 8, 35
(2015).
[121] Buvaneswari A. Ramanan, Lawrence M. Drabeck, Mark Haner, Nachi Nithi, Thierry
E. Klein, and Chitra Sawkar.
2013. Cacheability analysis of HTTP traffic in an operational LTE network. In
Proceedings of the IEEE Wireless
Telecommunications Symposium (WTS’13). IEEE, 1–8.
[122] Ammar Rayes and Salam Samer. 2016. Internet of Things From Hype to Reality:
The Road to Digitization. Springer.
[123] Ju Ren, Hui Guo, Chugui Xu, and Yaoxue Zhang. 2017. Serving at the edge: A
scalable IoT architecture based on
transparent computing. IEEE Netw. 31, 5 (2017), 96–105.
[124] Ju Ren, Yundi Guo, Deyu Zhang, Qingqing Liu, and Yaoxue Zhang. 2018.
Distributed and efficient object detection
in edge computing: Challenges and solutions. IEEE Netw. 32, 6 (2018), 137–143.
[125] Ju Ren, Yaoxue Zhang, Ruilong Deng, Ning Zhang, Deyu Zhang, and Xuemin Shen.
2016. Joint channel access and
sampling rate control in energy harvesting cognitive radio sensor networks. IEEE
Trans. Emerg. Top. Comput. 7, 1
(2019), 149–161. doi:10.1109/TETC.2016.2555806.
125:34
J. Ren et al.
[126] Morabito Roberto, Cozzolino Vittorio, Ding Aaron Yi, Beijar Nicklas, and Ott
Jörg. 2018. Consolidate IoT edge computing with lightweight virtualization. IEEE
Netw. 32, 1 (2018), 102–111.
[127] Tiago Gama Rodrigues, Katsuya Suto, Hiroki Nishiyama, and Nei Kato. 2017.
Hybrid method for minimizing service
delay in edge cloud computing through VM migration and transmission power control.
IEEE Trans. Comput. 66, 5
(2017), 810–819.
[128] Ola Salman, Sarah Abdallah, Imad H Elhajj, Ali Chehab, and Ayman Kayssi.
2016. Identity-based authentication
scheme for the Internet of Things. In Proceedings of the IEEE Symposium on
Computers and Communications (ISCC’16).
IEEE, 1109–1111.
[129] Mahadev Satyanarayanan. 2017. The emergence of edge computing. Computer 50, 1
(2017), 30–39.
[130] Mahadev Satyanarayanan, Paramvir Bahl, Ramón Cáceres, and Nigel Davies. 2009.
The case for VM-based cloudlets
in mobile computing. IEEE Perv. Comput. 8, 4 (2009), 14–23.
[131] Mahadev Satyanarayanan, Pieter Simoens, Yu Xiao, Padmanabhan Pillai, Zhuo
Chen, Kiryong Ha, Wenlu Hu, and
Brandon Amos. 2015. Edge analytics in the Internet of Things. IEEE Perv. Comput.
14, 2 (2015), 24–31.
[132] Min Sheng, Chao Xu, Junyu Liu, Jiongjiong Song, Xiao Ma, and Jiandong Li.
2016. Enhancement for content delivery
with proximity communications in caching enabled wireless networks: Architecture
and challenges. IEEE Commun.
Mag. 54, 8 (2016), 70–76.
[133] Adeshina Busari Sherif, Mumtaz Shahid, Al-Rubaye Saba, and Rodriguez
Jonathan. 2018. 5G millimeter-wave mobile
broadband: Performance and challenges. IEEE Commun. Mag. 56, 6 (2018), 137–143.
[134] Peng Shu, Fangming Liu, Hai Jin, Min Chen, Feng Wen, Yupeng Qu, and Bo Li.
2013. eTime: Energy-efficient transmission between cloud and mobile devices. In
Proceedings of the IEEE International Conference on Computer Communications
(INFOCOM’13). IEEE, 195–199.
[135] Javad Sohankar, Koosha Sadeghi, Ayan Banerjee, and Sandeep K. S. Gupta. 2015.
E-bias: A pervasive eeg-based
identification and authentication system. In Proceedings of the ACM International
Conference on Modeling, Analysis
and Simulation of Wireless and Mobile Systems (MSWiM’15). ACM, 165–172.
[136] Hong Song, Dacheng Wang, Jiaxin Wang, and Wenhao Zhang. 2017. TC-CCS: A
cooperative caching strategy in
mobile transparent computing system. In Proceedings of the IEEE IEEE International
Conference on Visual Communications and Image Processing (VCIP’17). IEEE, 1–4.
[137] Hongguang Sun, Matthias Wildemeersch, Min Sheng, and Tony QS Quek. 2015. D2D
enhanced heterogeneous cellular networks with dynamic TDD. IEEE Trans. Wireless
Commun. 14, 8 (2015), 4204–4218.
[138] Xiang Sun and Nirwan Ansari. 2017. Latency aware workload offloading in the
cloudlet network. IEEE Commun.
Lett. 21, 7 (2017), 1481–1484.
[139] Tarik Taleb, Konstantinos Samdanis, Badr Mada, Hannu Flinck, Sunny Dutta, and
Dario Sabella. 2017. On multiaccess edge computing: A survey of the emerging 5G
network edge architecture & orchestration. IEEE Commun.
Surv. Tutor. 19, 3 (2017), 1657–1681.
[140] Wenjuan Tang, Kuan Zhang, Ju Ren, Yaoxue Zhang, and Xuemin Sherman Shen.
2019. Flexible and efficient authenticated key agreement scheme for BANs based on
physiological features. IEEE Trans. Mobile Comput. 18, 4 (2019),
845–856.
[141] Yayuan Tang, Kehua Guo, and Biao Tian. 2018. A block-level caching
optimization method for mobile transparent
computing. Peer-to-Peer Network. Appl. 11, 4 (2018), 711–722.
[142] Stefano Traverso, Mohamed Ahmed, Michele Garetto, Paolo Giaccone, Emilio
Leonardi, and Saverio Niccolini. 2013.
Temporal locality in today’s content caching: Why it matters and how to model it.
ACM SIGCOMM Comput. Commun. Rev. 43, 5 (2013), 5–12.
[143] D. Trihinas, A. Tryfonos, M. D. Dikaiakos, and G. Pallis. 2018. DevOps as a
service: Pushing the boundaries of
microservice adoption. IEEE Internet Comput. 22, 3 (2018), 65–71.
[144] Luis M. Vaquero and Luis Rodero-Merino. 2014. Finding your way in the fog:
Towards a comprehensive definition
of fog computing. ACM SIGCOMM Comput. Commun. Rev. 44, 5 (2014), 27–32.
[145] Ricard Vilalta, Raluca Ciungu, Arturo Mayoral, Ramon Casellas, Ricardo
Martinez, David Pubill, Jordi Serra, Raul
Munoz, and Christos Verikoukis. 2016. Improving security in Internet of Things with
software defined networking.
In Proceedings of the IEEE Global Communications Conference (GLOBECOM’16). IEEE, 1–
6.
[146] Todd Vollmer and Milos Manic. 2014. Cyber-physical system security with
deceptive virtual hosts for industrial
control networks. IEEE Trans. Industr. Inf. 10, 2 (2014), 1337–1347.
[147] Serdar Vural, Ning Wang, Pirabakaran Navaratnam, and Rahim Tafazolli. 2017.
Caching transient data in internet
content routers. IEEE/ACM Trans. Netw. 25, 2 (2017), 1048–1061.
[148] Huaqun Wang, Zhiwei Wang, and Josep Domingo-Ferrer. 2017. Anonymous and
secure aggregation scheme in fogbased public cloud computing. Fut. Gener. Comput.
Syst. 78, 1 (2018), 712–719.
[149] Rui Wang, Xi Peng, Jun Zhang, and Khaled B. Letaief. 2016. Mobility-aware
caching for content-centric wireless
networks: Modeling and methodology. IEEE Commun. Mag. 54, 8 (2016), 77–83.
125:35
[150] Shuo Wang, Xing Zhang, Yan Zhang, Lin Wang, Juwo Yang, and Wenbo Wang. 2017.
A survey on mobile edge
networks: Convergence of computing, caching and communications. IEEE Access 5, 3
(2017), 6757–6779. DOI:https:
//doi.org/10.1109/ACCESS.2017.2685434
[151] Xiaofei Wang, Min Chen, Tarik Taleb, Adlen Ksentini, and Victor Leung. 2014.
Cache in the air: Exploiting content
caching and delivery techniques for 5G systems. IEEE Commun. Mag. 52, 2 (2014),
131–139.
[152] Xiumin Wang, Jin Wang, Xin Wang, and Xiaoming Chen. 2015. Energy and delay
tradeoff for application offloading
in mobile cloud computing. IEEE Syst. J. 11, 2 (2015), 858–867.
[153] Yanting Wang, Min Sheng, Xijun Wang, Liang Wang, and Jiandong Li. 2016.
Mobile-edge computing: Partial computation offloading using dynamic voltage
scaling. IEEE Trans. Commun. 64, 10 (2016), 4268–4282.
[154] Shinae Woo, Eunyoung Jeong, Shinjo Park, Jongmin Lee, Sunghwan Ihm, and
KyoungSoo Park. 2013. Comparison
of caching strategies in modern cellular backhaul networks. In Proceedings of the
ACM International Conference on
Mobile Systems, Applications, and Services (Mobisys’13). ACM, 319–332.
[155] Huaming Wu, Yi Sun, and Katinka Wolter. 2018. Energy-efficient decision
making for mobile cloud offloading. IEEE
Trans. Cloud Comput. (2018). DOI:https://doi.org/10.1109/TCC.2018.2789446
[156] Changqiao Xu, Tianjiao Liu, Jianfeng Guan, Hongke Zhang, and Gabriel-Miro
Muntean. 2013. CMT-QA: Qualityaware adaptive concurrent multipath data transfer in
heterogeneous wireless networks. IEEE Trans. Mobile Comput.
12, 11 (2013), 2193–2205.
[157] George Xylomenos, Christopher N. Ververidis, Vasilios A. Siris, Nikos Fotiou,
Christos Tsilopoulos, Xenofon Vasilakos, Konstantinos V. Katsaros, and George C.
Polyzos. 2014. A survey of information-centric networking research.
IEEE Commun. Surv. Tutor. 16, 2 (2014), 1024–1049.
[158] Jianxiao Yang, Benoît Geller, and Stéphanie Bay. 2011. Bayesian and hybrid
Cramér–Rao bounds for the carrier
recovery under dynamic phase uncertain channels. IEEE Trans. Sign. Process. 59, 2
(2011), 667–680.
[159] Lei Yang, Jiannong Cao, Hui Cheng, and Yusheng Ji. 2015. Multi-user
computation partitioning for latency sensitive
mobile cloud applications. IEEE Trans. Comput. 64, 8 (2015), 2253–2266.
[160] M. Yannuzzi, R. Milito, R. Serral-Gracia, D. Montero, and M. Nemirovsky.
2014. Key ingredients in an IoT recipe:
Fog computing, cloud computing, and more fog computing. In Proceedings of the IEEE
International Workshop on
Computer Aided Modeling and Design of Communication Links and Networks (CAMAD’14).
325–329. DOI:https://doi.
org/10.1109/CAMAD.2014.7033259
[161] Shanhe Yi, Cheng Li, and Qun Li. 2015. A survey of fog computing: Concepts,
applications and issues. In Proceedings
of the 2015 ACM Workshop on Mobile Big Data (Mobidata’15). ACM, 37–42.
[162] Changsheng You and Kaibin Huang. 2016. Multiuser resource allocation for
mobile-edge computation offloading.
arXiv preprint arXiv:1604.02519 (2016).
[163] Changsheng You, Kaibin Huang, Hyukjin Chae, and Byoung-Hoon Kim. 2017.
Energy-efficient resource allocation
for mobile-edge computation offloading. IEEE Trans. Wireless Commun. 16, 3 (2017),
1397–1411.
[164] Engin Zeydan, Ejder Bastug, Mehdi Bennis, Manhal Abdel Kader, Ilyas Alper
Karatepe, Ahmet Salih Er, and
Mérouane Debbah. 2016. Big data caching for networking: Moving from cloud to edge.
IEEE Commun. Mag. 54,
9 (2016), 36–42.
[165] Guanglin Zhang, Wenqian. Zhang, Yu. Cao, Demin Li, and Lin Wang. 2018.
Energy-delay tradeoff for dynamic
offloading in mobile-edge computing system with energy harvesting devices. IEEE
Trans. Industr. Inf. 14, 10 (2018),
4642–4655.
[166] Ke Zhang, Supeng Leng, Yejun He, Sabita Maharjan, and Yan Zhang. 2018.
Cooperative content caching in 5G
networks with mobile edge computing. IEEE Wireless Commun. 25, 3 (2018), 80–87.
[167] Ke Zhang, Supeng Leng, Yejun He, Sabita Maharjan, and Yan Zhang. 2018. Mobile
edge computing and networking
for green and low-latency Internet of Things. IEEE Commun. Mag. 56, 5 (2018), 39–
45.
[168] Ke Zhang, Yuming Mao, Supeng Leng, Yejun He, and Yan Zhang. 2017. Mobile-edge
computing for vehicular networks: A promising network paradigm with predictive off-
loading. IEEE Vehic. Technol. Mag. 12, 2 (2017), 36–44.
[169] Ke Zhang, Yuming Mao, Supeng Leng, Quanxin Zhao, Longjiang Li, Xin Peng, Li
Pan, Sabita Maharjan, and Yan
Zhang. 2016. Energy-efficient offloading for mobile edge computing in 5G
heterogeneous networks. IEEE Access 4,
8 (2016), 5896–5907.
[170] Meng Zhang, Hongbin Luo, and Hongke Zhang. 2015. A survey of caching
mechanisms in information-centric
networking. IEEE Commun. Surv. Tutor. 17, 3 (2015), 1473–1499.
[171] Weiwen Zhang, Yonggang Wen, and Dapeng Oliver Wu. 2015. Collaborative task
execution in mobile cloud computing under a stochastic wireless channel. IEEE
Trans. Wireless Commun. 14, 1 (2015), 81–93.
[172] Yaoxue Zhang. 2004. Transparence computing: Concept, architecture and
example. Chin. J. Electr. 32, S1, Article 169
(2004), 5 pages.
[173] Yaoxue Zhang, Kehua Guo, Ju Ren, Yuezhi Zhou, Jianxin Wang, and Jianer Chen.
2017. Transparent computing: A
promising network computing paradigm. Comput. Sci. Eng. 1, 2 (2017), 7–20.
125:36
J. Ren et al.
[174] Yaoxue Zhang, Ju Ren, Jiagang Liu, Chugui Xu, Hui Guo, and Yaping Liu. 2017.
A survey on emerging computing
paradigms for big data. Chin. J. Electr. 26, 1 (2017), 1–12.
DOI:https://doi.org/10.1049/cje.2016.11.016
[175] Yaoxue Zhang, Laurencetianruo Yan, Yuezhi Zhou, and Wenyuan Kuang. 2013.
Information security underlying
transparent computing: Impacts, visions and challenges. Web Intell. Agent Syst. 8,
2 (2013), 203–217.
[176] Yaoxue Zhang and Yuezhi Zhou. 2006. Transparent computing: A new paradigm for
pervasive computing. In Proceedings of the 3rd International Conference on
Ubiquitous Intelligence and Computing (UIC’06). Springer-Verlag, Berlin,
1–11. DOI:https://doi.org/10.1007/11833529_1
[177] Yaoxue Zhang and Yuezhi Zhou. 2007. 4VP: A novel meta OS approach for
streaming programs in ubiquitous computing. In Proceedings of the IEEE
International Conference on Advanced Information Networking and Applications
(AINA’07). IEEE, 394–403.
[178] Yaoxue Zhang and Yuezhi Zhou. 2013. Transparent computing: Spatio-temporal
extension on von Neumann architecture for cloud services. Tsing. Sci. Technol. 18,
1 (Feb. 2013), 10–21.
[179] Tianchu Zhao, Sheng Zhou, Xueying Guo, Yun Zhao, and Zhisheng Niu. 2015. A
cooperative scheduling scheme
of local cloud and internet cloud for delay-aware mobile cloud computing. In
Proceedings of the IEEE GLOBECOM
Workshops. IEEE, 1–6.
[180] Yun Zhao, Sheng Zhou, Tianchu Zhao, and Zhisheng Niu. 2015. Energy-efficient
task offloading for multiuser mobile
cloud computing. In Proceedings of the IEEE International Conference on Cognitive
Computing (ICCC’15). IEEE, 1–5.
[181] Zhongyuan Zhao, Mugen Peng, Zhiguo Ding, Wenbo Wang, and H Vincent Poor.
2016. Cluster content caching: An
energy-efficient approach to improve quality of service in cloud radio access
networks. IEEE J. Select. Areas Commun.
34, 5 (2016), 1207–1221.
[182] Cong Zuo, Jun Shao, Guiyi Wei, Mande Xie, and Min Ji. 2018. CCA-secure ABE
with outsourced decryption for fog
computing. Future Gener. Comput. Syst. 78, 1 (2018), 730–738.
Received January 2019; revised July 2019; accepted September 2019

Quiz 01 Ren

Uploaded by

Copyright:

Available Formats

Quiz 01 Ren

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quiz 01 Ren

Uploaded by

Copyright:

Available Formats

A Survey on End-Edge-Cloud Orchestrated Network

Computing Paradigms: Transparent Computing, Mobile

Fig. 1. An illustration of the TC architecture.

Fig. 2. An illustration of TC-based IoT system.

Fig. 3. A typical BaaS architecture.

Fig. 4. Architecture of mobile edge computing.

Fig. 5. Architecture of fog computing.

Fig. 6. Architecture of cloudlet-based computing [131].

Location for Computing

works design stochastic optimization approaches to improve the adaptability of ad

Discussion of the Similarities and Differences

The common features of the aforementioned emerging computing paradigms are to

software resources, including OS and various application software (Apps), and to

COMPUTATION OFFLOADING FOR EMERGING COMPUTING PARADIGMS

One of the main targets of end-edge-cloud orchestrated computing paradigms is to

Joint Optimization of Energy Consumption and Delay

To improve the performance of service provisioning in the end-edge-cloud

respectively. In Reference [171], the authors partition an application into a

investigates the computation offloading in an interference environment. First, the

CACHING FOR EMERGING COMPUTING PARADIGMS

In the emerging end-edge-cloud orchestrated architecture, nodes at different levels

Placing Caching Units

In the end-edge-cloud orchestrated architecture, edge devices should be able to

tremendous traffic on Internet. Moreover, a large number of users in mobile

It is crucial to decide what caching strategies to be adopted for caching at the

facilitates the context-aware delivery by the deployment of in-network caching

5.2 Service Security and Privacy

Reference [91] proposes a privacy-preserving data aggregation scheme to aggregate

FUTURE RESEARCH DIRECTIONS

By fully exploiting the resources of end devices and the nearby

Lightweight Virtualization for Network Computing

Fig. 7. APIs for end-edge-cloud orchestrated computing systems.

carefully redesigned to address the security risk. In addition, if the software of

Lightweight and Distributed Machine Learning for Realtime Intelligence

Received January 2019; revised July 2019; accepted September 2019

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.