Digital Communications and Networks: Yuan Ai, Mugen Peng, Kecheng Zhang

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Digital Communications and Networks 4 (2018) 77–86

Contents lists available at ScienceDirect

Digital Communications and Networks


journal homepage: www.keaipublishing.com/en/journals/digital-communications-and-networks/

Edge computing technologies for Internet of Things: a primer


Yuan Ai, Mugen Peng *, Kecheng Zhang
Key Laboratory of Universal Wireless Communications for Ministry of Education, Beijing University of Posts and Telecommunications, China

A R T I C L E I N F O A B S T R A C T

Keywords: With the rapid development of mobile internet and Internet of Things applications, the conventional centralized
Internet of Things (IoT) cloud computing is encountering severe challenges, such as high latency, low Spectral Efficiency (SE), and non-
Mobile edge computing adaptive machine type of communication. Motivated to solve these challenges, a new technology is driving a
Cloudlets trend that shifts the function of centralized cloud computing to edge devices of networks. Several edge computing
Fog computing technologies originating from different backgrounds to decrease latency, improve SE, and support the massive
machine type of communication have been emerging. This paper comprehensively presents a tutorial on three
typical edge computing technologies, namely mobile edge computing, cloudlets, and fog computing. In particular,
the standardization efforts, principles, architectures, and applications of these three technologies are summarized
and compared. From the viewpoint of radio access network, the differences between mobile edge computing and
fog computing are highlighted, and the characteristics of fog computing-based radio access network are discussed.
Finally, open issues and future research directions are identified as well.

1. Introduction (i.e., things) [5]. IoT's features, including ultra-largescale network of


things, device and network level heterogeneity, and large numbers of
Over the past decades, cloud computing has been greatly developed events generated by these things, will make the development of diverse
and applied owing to its high cost-efficiency and flexibility achieved applications and services a very challenging task [6]. These requirements
through consolidation, in which computing, storage, and network man- are becoming difficult to accomplish in the IoTþ cloud scenario. IoT
agement functions work in a centralized manner. With the rapid devel- applications generate enormous amounts of data by IoT sensors. Big data
opment of mobile internet and Internet of Things (IoT) applications, the are subsequently analyzed to determine reactions to events or to extract
existing centralized cloud computing architecture is encountering severe analytics or statistics. However, sending all the data to the cloud will
challenges. Mobile devices connected to distant centralized cloud servers require prohibitively high network bandwidth. Recent research efforts
try to obtain sophisticated applications, which impose additional load on are investigating on how to effectively exploit capabilities at the edge of
both Radio Access Networks (RANs) and backhaul networks and result in networks to support the IoT and its requirements [7]. In edge computing,
high latency [1]. In addition, with the explosive growth in various access the massive data generated by different types of IoT devices can be
devices and end-user demands, IoT is driving a digital transformation in processed at the network edge instead of transmitting them to the
all aspects of the current modern life [2]. It is estimated by Cisco that the centralized cloud infrastructure owing to bandwidth and energy con-
number of devices connected to IoT will become 50 billion by 2020 [3]. sumption concerns. Edge computing can provide services with faster
The emerging IoT introduces new challenges, such as stringent latency, response and greater quality, in comparison with cloud computing. Edge
capacity constraints, resource-constrained devices, uninterrupted ser- computing is more suitable to be integrated with IoT to provide efficient
vices with intermittent connectivity, and enhanced security, which and secure services for a large number of end-users, and edge
cannot be adequately addressed by the centralized cloud computing ar- computing-based architecture can be considered for the future IoT
chitecture [4]. An advanced cloud computing paradigm that breaks infrastructure [8].
through the centralized architecture and alleviates the capacity and la- Recently, nascent technologies and applications are driving a trend in
tency constraints is urgently required to cope with these challenges. the computing and communication landscape that shifts the function of
IoT refers to the interaction and communication between billions of centralized cloud computing into the edge devices of networks [9].
devices that produce and exchange data related to real-world objects Software Defined Networking (SDN) and the associated concept of

* Corresponding author.
E-mail addresses: [email protected] (Y. Ai), [email protected] (M. Peng), [email protected] (K. Zhang).

https://doi.org/10.1016/j.dcan.2017.07.001
Received 24 April 2017; Received in revised form 28 June 2017; Accepted 2 July 2017
Available online 8 July 2017
2352-8648/© 2017 Chongqing University of Posts and Telecommunications. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-
NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

Network Function Virtualization (NFV) are proposed as emerging solu- Table 1


tions for future networks [10]. In particular, NFV enables edge devices to Summary of existing survey articles on edge computing.
provide computing services and operate network functions by creating Aspects Survey Contributions
multiple Virtual Machines (VMs). Moreover, ultra-low latency is identi- papers
fied as one of the major requirements of 5th Generation (5G) RANs [11]. Cloudlet [15] A new cloudlet-based architecture for
To decrease the latency, mobile operators are prone to deploying the overcoming the technical obstacles in mobile
application and content at the edge of networks. Meanwhile, operators computing.
can open the edge devices of RANs to third-party partners, allowing them [24] An open ecosystem based on the concept of
cloudlets supporting many exciting mobile
to rapidly deploy innovative applications and content toward mobile applications.
subscribers, enterprisers, and other vertical segments [12]. Although the Mobile edge [12] An overview of MEC definition, architectural
computing capabilities of wearable watches, smart phones, and other IoT computing blueprint, requirements, and challenges of MEC
devices have been significantly improved, they are still constrained by as well as the objectives of the MEC initiative.
[13] A comprehensive survey of the state-of-the-art
the fundamental challenges, such as memory size, battery life, and heat
MEC research focusing on joint radio and
dissipation. Mobile devices need to extend battery lifetime by offloading computational resource management.
energy-consuming computation of applications to the edge of networks [41] A comprehensive survey of major use cases and
[13]. reference scenarios, current advancement in
Motivated to efficiently provide the massive machine type of standardization of MEC, and research on
computation offloading.
communication, ultra-reliable low-latency communication, and high Fog computing [4] A summary of the opportunities and challenges
Spectral Efficiency (SE), industry investment and research interest of fog computing focusing primarily on the
focused on the edge of networks have grown dramatically [14]. To networking context of IoT.
support low-latency requirements for resource-intensive applications, a [54] An overview of Fog computing definition,
reference architecture, use cases and challenges
new architectural element called cloudlets has been proposed in
for fog computing as well as the future research
Ref. [15]. In order to accelerate the development of the ecosystem based and work.
on cloudlets, the Open Edge Computing (OEC) initiative has been Comprehensive [9] An overview of edge computing definition,
launched in June 2015 by Vodafone, Intel, and Huawei companies in surveys origin and background, challenges, and
partnership with Carnegie Mellon University. Similarly, Nokia Networks applications. Discussions of the future research
directions of edge computing.
company introduced a computing platform in 2013, which is integrated
[14] A comprehensive comparison of three
to the base station [16]. The original concept that applications and ser- approaches: fog computing, MEC and Cloudlet.
vices are executed at the edge of networks has evolved gradually. In Discussions of further work and research in
September 2014, a new Industry Specification Group (ISG) was proposed order to get concepts like Fog, MEC and
Cloudlets adopted by industry.
to be set up in European Telecommunications Standards Institute (ETSI)
This A comprehensive tutorial of three state-of-the-
to allow the creation of industry specifications for Mobile Edge article art edge computing technologies, namely MEC,
Computing (MEC), which has been supported by Huawei, IBM, Intel, cloudlets, and fog computing. A comparison of
Nokia Networks, NTT DoCoMo, Vodafone, and other companies [17]. In standardization efforts, principles,
the MEC World Congress 2016, MEC ISG has renamed mobile edge architectures, and applications for these three
technologies. The difference between mobile
computing as multi-access edge computing in order to reflect the growing
edge computing and fog computing from the
interests from non-cellular operators. It is anticipated that multi-access viewpoint of RANs.
edge computing will take effect starting from the end of 2017 [18].
Another technology similar to edge computing is known as fog
computing, which was initiated by Cisco in 2012 [19]. To accelerate the standardization efforts, applications, and system architecture of fog
adoption of fog computing, the OpenFog Consortium has been founded computing, and compares MEC and fog computing-based RANs (F-
by ARM, Cisco, Dell, Intel, Microsoft and Princeton University in RANs). The open issues and challenges are discussed in Section 5, prior to
November 2015. the conclusion in Section 6.
As there are three typical edge computing solutions for IoT, it is
necessary to highlight the differences among them, and the development 2. Cloudlet: the edge of internet
of a framework for edge computing technologies with reference to
background, system architecture, and key techniques is challenging. The One of the critical challenges in cloud computing is the end-to-end
existing tutorial articles on edge computing are presented in Table 1. The responsiveness between a mobile device and associated cloud [20]. To
scope of these existing publications provided neither an in-depth dis- address this challenge, the cloudlet, which is a mobility-enhanced
cussion nor a comparison in terms of improving SE, decreasing latency, small-scale cloud Data Center (DC) that is located at the edge of the
and supporting the massive machine type of communication from the internet, is proposed. A cloudlet is a trusted, resource-rich computer or
perspectives of the overall system architecture and key techniques. It is cluster of computers that are well-connected to the internet and available
critical, therefore, to show a tutorial on edge computing by presenting a for use by nearby mobile devices [15].
comprehensive review framework. Considering the important aspects The main purpose of the cloudlet is supporting resource-intensive and
and active research activities of edge computing, a tutorial of system interactive mobile applications by providing powerful computing re-
architectures, key techniques, and application characteristics is pre- sources to mobile devices with lower latency. User Equipments (UEs) can
sented in this paper to promote research and commercial success of edge access the computing resources in the nearby cloudlet through a one-hop
computing. Additionally, given the extensiveness of the research area, high-speed wireless local area network. Cloudlets represent the middle
more open issues and challenging works on both transforming the well- tier of the 3-tier hierarchy architecture (mobile device layer, cloudlet
established infrastructure of fog computing for 5G RANs that are neces- layer, and cloud layer) to achieve crisp response time. There are few but
sary to perform elaborative investigations in the future, are introduced as important differentiators between cloud and cloudlet:
well.
The rest of this paper is organized as follows: Section 2 presents a 1) Compared to the cloud DC, a cloudlet needs to be much more agile in
review on the principles and applications of cloudlets. The standardiza- its provisioning because the association with mobile devices is highly
tion efforts, applications, architecture, and key technologies of mobile dynamic with considerable churn due to user mobility;
edge computing are presented in Section 3. Section 4 summarizes the

78
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

2) To support user mobility, a VM handoff technology needs to be used [26]. By serving as physically proximate representatives of the cloud that
to seamlessly migrate the offloaded services on the first cloudlet to the are unavailable owing to failures or cyberattacks, cloudlets can improve
second cloudlet as a user moves away from the currently associated robustness and availability in hostile environment [27]. Moreover,
cloudlet; cloudlets can enable mobile access to the huge legacy world of
3) Because cloudlets are small DCs distributed geographically, a mobile Windows-based desktop applications. A VM encapsulating the personal
device first has to discover, select, and associate with the appropriate desktop environment of a user is run on a cloudlet, and the user connects
cloudlet among multiple candidates before it starts provisioning. to it through a remote desktop protocol. Thus, users can use
Windows-based desktop applications on a mobile device such as an
Android tablet that is typically ARM-based.
To drive the development of software ecosystems surrounding
2.1. Principle of cloudlet
cloudlets, the OEC initiative has been launched, synchronizing work with
other efforts by ETSI ISG MEC and OPNFV. In addition, Carnegie Mellon
Mobile users exploit the VM technology to rapidly instantiate a
University has implemented an open source platform called Open-
customized software on a nearby cloudlet. Generally, the customization
Stackþþ that is a derivative of the widely used OpenStack platform for
of a base system specialized for a certain application is small. If the base
cloud computing, which extends the functionality of Openstack to sup-
VM exists on the cloudlet, only its difference relative to the desired
port cloudlets. Some key technologies such as cloudlet discovery, just-in-
custom VM, called a VM overlay, needs to be transferred. The approach of
time provisioning and VM hand-off, have been implemented to be
using VM overlays to provisioning cloudlets is called VM synthesis [21].
available as open source [28].
Cloudlets use the approach of VM synthesis for rapid provisioning and
VM handoff. Fig. 1 shows the relevant steps of dynamic VM synthesis. A
mobile device delivers a small VM overlay to the cloudlet that already 3. Mobile edge computing: the edge of mobile network
possesses the base VM from which the overlay is derived (the delivery
can be either from the cloud or from the storage on the mobile device). MEC is identified as a key enabler for IoT and for mission-critical,
The cloudlet decompresses the overlay, applies it to the base VM to vertical solutions, and is recognized as one of the key architectural
derive the launched VM, and then creates a VM instance from it. The concepts and technologies. The concept of MEC was defined by ETSI as a
mobile device can begin performing offload operations at this instance. new technology that provides an IT service environment and cloud-
At the end of the session, the instance is destroyed, but the launched VM computing capabilities at the edge of the mobile network, within the
image can be retained in a persistent cache for future sessions. To retain RAN and in close proximity to mobile subscribers [12]. ETSI has pub-
some training data for future offload sessions, the cloudlet generates a lished a white paper on MEC, and MEC has been considered a key
VM residue that can be sent back to the mobile device and incorporated emerging technology, which is important for future generation networks
into its overlay. Experimental results show that cloudlets can decrease [17].
the response time by 51% and reduce energy consumption by up to 42%
in a mobile device compared to cloud offload [22]. The VM handoff is
3.1. MEC standardization
responsible for seamlessly transferring VM-encapsulated execution to an
optimal offload site as users move. Validation experiments confirm that
ETSI has established the ISG on MEC to develop a standardized, open
the resulting mechanism is a promising technique for enabling user
environment that will allow efficient and seamless integration of third-
mobility with low end-to-end latency applications [23].
party applications across multi-vendor platforms in December 2014.
Until January 2017, MEC ISG has released six specifications, one of
2.2. Application of cloudlet which provides a glossary of terms related to the conceptual, architec-
tural, and functional elements of MEC [29]. The purpose of that speci-
An open ecosystem based on the concept of cloudlets supports and fication is to enable the consistent use of terminology within ETSI MEC
enables many exciting mobile applications that are both compute- specifications and, beyond the ISG, more widely in the industry. Another
intensive and latency-sensitive [24]. For example, by leveraging a low specification provides technical requirements enabling interoperability
end-to-end latency, the real-time interaction can be implemented on and deployment and describes example use cases and their technical
wearable cognitive assistance [25]. By real-time data analysis at the edge benefits [30]. The other specification provides a framework and refer-
of the internet, cloudlets can reduce ingress bandwidth into the cloud ence architecture to enable mobile edge applications to run efficiently

Fig. 1. Dynamic VM synthesis.

79
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

and seamlessly in a mobile network [31]. Moreover, the fourth specifi- require low latency and a high rate of data processing in order to provide
cation in MEC ISG introduces a number of service scenarios that would the correct information depending on the location of the user. The pro-
benefit from the MEC technology [32]. The Proof of Concept (PoC) cessing of data can be performed on a local MEC server rather than on a
framework specification defines a framework to coordinate and promote centralized server to provide a perfect user experience. In addition, a
multi-vendor PoC projects illustrating key aspects of MEC technology vehicular delay-tolerant network-based smart grid data management
[33]. PoCs are essential to demonstrate MEC as feasible and valuable, to scheme that leverages the mobile edge computing paradigm was pro-
validate the specifications that are being developed, to demonstrate use posed in Ref. [37]. The IoT generates additional messaging on telecom-
cases, and ultimately to help develop a diverse and open MEC ecosystem. munication networks, and requires gateways to aggregate the messages
The last specification describes various metrics, which can potentially be and ensure low latency and security. A new architecture by leveraging
improved through deploying a service on a MEC platform, such as la- MEC to collect, classify, and analyze the IoT data streams was introduced
tency, energy efficiency, network throughput, system resource footprint, in Ref. [38]. The MEC server is responsible for managing various pro-
and quality [34]. Furthermore, the last specification also describes the tocols, distribution of messages, and for processing of analytics. The MEC
best practices for measuring such performance metrics. ETSI has environment creates a new value chain and an energized ecosystem,
announced six different MEC PoCs in September 2016, which have been which in turn creates new opportunities for mobile operators and
accepted in MEC World Congress in Munich and have contributed to application and content providers.
strengthen the strategic planning and decision-making of organizations,
and help to identify which MEC solutions may be viable in the network.
3.3. System architecture of MEC
MEC ISG is now working on 9 new studies related to MEC APIs, man-
agement interfaces, and essential platform functionality. In addition, the
As shown in Fig. 2, the MEC reference architecture, as described by
MEC in an NFV environment is emerging on an end-to-end mobility. The
ETSI [31], enables the implementation of MEC applications as
NFV platform may be dedicated to MEC or shared with other network
software-only entities that run on the MEC host. The mobile edge plat-
functions or applications. MEC exploits the NFV management and
form offers the essential environment and functionality required to run
orchestration entities and interfaces as much as possible.
the MEC application. MEC applications are running as VM on top of the
virtualization infrastructure, and can interact with the mobile edge
3.2. Application of MEC platform to perform certain support procedures related to the life-cycle of
the application. Furthermore, the virtualization infrastructure includes a
Owing to its advanced features, such as low latency, proximity, high data plane that executes the traffic rules received by the mobile edge
bandwidth, and real-time insight into radio network information and platform and routes the traffic among applications, local networks, and
location awareness, MEC enables a large number of new types of appli- external networks. The MEC host level management comprises the mo-
cations and services for multiple sectors, such as consumer, enterprise, bile edge platform manager and the virtualization infrastructure man-
and health. In particular, MEC is deemed to be a promising solution for ager. The former manages the life cycle of applications and the
handling video streaming services in the context of smart cities [35]. application rules and requirements including service authorizations,
Video streams from monitoring devices are locally processed and traffic rules, DNS configuration, and resolving conflicts. The latter is
analyzed at a MEC server to extract meaningful data from video streams. responsible for allocating, managing, and releasing visualized (compute,
The valuable data can be transmitted to the application server to reduce storage, and networking) resources of the virtualization infrastructure.
core network traffic. Augmented Reality (AR) mobile applications have The operations support system receives request by a user application
inherent collaborative properties in terms of data collection in the uplink, via a life-cycle management proxy, or by operators' third-party customers
computing at the edge, and data delivery in the downlink [36]. AR data via the customer-facing service portal. The operation support system

Fig. 2. Mobile edge system reference architecture.

80
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

decides whether the requests are granted or not. The granted requests are algorithm achieves an excellent computation offloading performance and
forwarded to the MEC orchestrator for further processing. The MEC scales well as the user size increases. The energy-efficient resource allo-
orchestrator is the core functionality, as it maintains an overall view cation problems for computation offloading have been researched in
based on the deployed MEC hosts, available resources, available MEC Ref. [45]. In addition, some efforts have been focused on the joint opti-
services, and topology. For reasons of performance, costs, scalability, and mization of radio and computational resources, aimed at minimizing en-
operator preferred deployments, MEC supports different deployment ergy consumption under latency and power budget constraints [46].
scenarios [30], such as at the cellular macro base station (eNodeB) site, at When the UE performs a handover to another cell, it is important to
the 3G Radio Network Controller (RNC) site, at a multi-radio access guarantee the service continuity and QoS requirements [47]. Several
technology cell aggregation site, and at an aggregation point (which may MEC applications expect to continue serving the UE after a location
also be at the edge of the core network, e.g., in a distributed DC). A change. If the UE forwards the computation to the MEC, VM migration
network planning problem on determining where to optimally install the needs to be initiated to provide continuity of the service. VM migration is
MEC servers among the available sites to find a tradeoff between a service included in many hyper-visors to move VMs from one physical
installation costs and Quality of Service (QoS) has been explored in machine to another, commonly within a DC [48]. A crucial part is to
Ref. [39]. decide whether VM migration depends on the service type and re-
quirements, content size, and user class [49]. Note that an ongoing video
3.4. Key technologies of MEC service with strict QoS requirements may be migrated, and delay-sensi-
tive measurement tasks for an emergency warning machine type
The key technologies of MEC including computation offloading and communication services are always migrated to the optimal MEC server.
mobility management are illustrated in Fig. 3. Computation offloading is a On one hand, a decision must be made indicating whether the service
procedure that migrates resource-intensive computations from a mobile must be fully or partially migrated, while considering the VM migration
device to the resource-rich nearby infrastructure [40]. Although mobile cost including the time required for the VM migration and backhaul re-
devices are constrained by computing capabilities, battery life, and heat sources consumption because of traffic exchanges among computing
dissipation, MEC enables running new sophisticated applications at UEs nodes. On the other hand, the benefit of VM migration is to reduce the
by offloading energy-consuming computations of the applications to the service delay. Meanwhile, backhaul resources do not have to be allocated
MEC server. An important part of computation offloading is to decide for the transmission of computation results back to the UE. A profit
whether to offload or not, whether full or partial offloading is applicable, maximization avatar placement strategy was proposed in order to opti-
and what and how the computation could be offloaded. The offloading mize the tradeoff between the migration gain and the migration cost by
decision depends on the application model, which can be classified ac- selectively migrating the VM to their optimal locations [50]. In addition,
cording to three criteria [41]. The first criterion is whether the application a prediction technology used for the dynamic VM placement and to find
contains non-offloadable parts that cannot be offloaded (e.g., user input, the most suitable communication path according to the expected users'
camera, or acquired position that need to be executed at UEs). Second, movement has been explored in Ref. [51]. Comparing to state-of-the-art
there is no way to estimate the amount of data to be processed for some approaches, the proposed algorithm reduces the offloading delay by a
continuous-execution applications. The third criterion is a mutual de- value between 10% and 66%.
pendency of individual parts to be processed. Generally, UE needs to be
composed of a code profiler, system profilers, and decision engine to 4. Fog computing: edge work with cloud
manage the offloading process [42]. The code profiler is responsible for
managing what to offload depending on the application type and code/- The OpenFog Consortium was founded to drive industry and aca-
data partitioned. System profilers are in charge of monitoring multiple demic leadership in fog computing architecture, testbed development,
parameters, such as available bandwidth, data size to transmit, and energy and a variety of inter-operability and composability deliverables that
to execute the code. These parameters influence when to offload. The seamlessly leverage cloud and edge architectures to enable end-to-end
decision engine determines whether to offload or not. Computation off- IoT scenarios [52]. OpenFog Consortium published a white paper on
loading decision algorithms have been investigated and compared fog computing in February 2016, in which the consortium’s approach to
comprehensively in Ref. [41]. The majority of algorithms aim to minimize an open fog computing architecture (OpenFog architecture) has been
the energy consumption at the mobile device, subject to the execution outlined [53]. The OpenFog Consortium defines fog computing as a
delay acceptable by the offloaded application, or to find an optimal system-level horizontal architecture that distributes resources and ser-
tradeoff between these two metrics. Numerical results demonstrate that vices of computing, storage, control and networking anywhere along the
MEC can improve energy efficiency by computation offloading in het- continuum from the cloud to things. Fog computing is different from edge
erogeneous networks [43]. A game theoretic approach was proposed for a computing and provides tools for distributing, orchestrating, managing,
computation offloading decision-making problem among multiple mobile and securing resources and services across networks and between devices
device users in Ref. [44]. Numerical results show that the proposed that reside at the edge. Edge architecture places servers, applications, and

Fig. 3. Offloading and mobility management.

81
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

small clouds at the edge. Fog jointly works with the cloud, while edge is 4.2. Application of fog computing
defined by the exclusion of cloud.
An open architecture based on fog computing enable interoperability
4.1. Standardization of fog computing in IoT, 5G, AI, tactile internet, virtual reality, and other complex data and
network intensive applications. IoT applications generate unprecedented
The fog computing standardization is mainly charged by the OpenFog amounts of data that can be useful in many ways [55]. Based on this sit-
Consortium, whose objective is to influence standards bodies to create uation, fog nodes can be used to carry out data mining and data analysis on
standards so that IoT systems at the edge can inter-operate securely with a large volume of multi-modal and heterogeneous data from various
other edge and cloud services in a friction-free environment. The sensor devices and other IoT devices to achieve real time and fast pro-
OpenFog Consortium has set up six working groups, namely architecture cessing for decision making [56]. A hierarchical fog computing architec-
working group, communications working group, manageability working ture for big data analysis in smart cities was introduced in Ref. [57].
group, security working group, software infrastructure working group, Meanwhile, a prototypical system for smart pipeline monitoring was
and testbed working group. These working groups evaluate, classify, and constructed to experimentally evaluate the performance of this architec-
recommend standards, practices, and technologies that are appropriate ture. Experimental results demonstrate the feasibility of the system's
for the OpenFog architecture to address corresponding challenges. city-wide implementation in future smart cities scenario. A
The OpenFog Consortium announces the release of the OpenFog privacy-preserving protocol for enhancing security in vehicular crowd
Reference Architecture in February 2017, which is a universal technical sensing-based road surface condition monitoring system using fog
framework designed to enable the data-intensive requirements of IoT, computing was proposed in Ref. [58]. The fog computing-based face
5G, and Artificial Intelligence (AI) applications [54]. This architecture is identification and resolution framework have been explored to solve some
the baseline to develop an open architecture fog computing environment, security and privacy issues [59]. In addition, because the fog is localized,
which creates a roadmap and is the first step in creating standards for fog new services that require mobile networks supporting high data rates and
computing. The OpenFog Consortium will establish detailed guidance, low latency become possible, such as virtual reality. The vehicle can be
interface with standards organizations such as IEEE on recommending employed as the fog node to attain optimum utilization of these vehicular
standards, and specifying APIs for key interfaces in the reference communications and computational resources [60]. The mobile fog node
architecture. can communicate with other fog nodes or provide services including
The structural aspects and perspectives of the reference architecture, infotainment, advanced driver assistance systems, autonomous driving,
which is used as a common baseline for achieving a multi-vendor inter- collision avoidance, and navigation. Emergency, health care, and other
operable fog computing ecosystem, are illustrated in Fig. 4. It is a com- latency-sensitive and security-/privacy-sensitive services require fog
posite of multiple views to address stake-holders in the fog computing nodes to be executed between the underlying nodes and the distant cloud
value chain, such as software view, system view, and node view. The [61]. Extensive experimental results validate that fog computing sup-
node view is the lowest level view, which includes the protocol porting medical cyber-physical system can improve the cost efficiency
abstraction layer and sensors, actuators, and control. The system view is significantly by jointly considering base station association, task distri-
composed of one or more node views coupled with other components to bution, and VM deployment [62]. Fog computing provides business value
create a platform. Software view comprises the top three layers that sit on for some applications that require real-time decision making, low latency,
top of the platform hardware layer. The software is running on fog improved security, and are network-constrained.
platforms to satisfy a use case requirement. Five cross-cutting perspec-
tives are employed throughout fog computing implementations, namely
(1) performance and scale perspective, (2) security perspective, (3) 4.3. System architecture of fog computing
manageability perspective, (4) data analytic and control perspective, and
(5) IT business and cross-fog applications perspective. A typical hierarchical architecture based on fog computing is shown
in Fig. 5. From the functional point of view, a fog node has several
functions, including networking, computing, accelerating, storing, and

Fig. 4. The OpenFog reference architecture description with perspectives.

82
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

Fig. 5. A typical hierarchical architecture based on fog computing.

control. Fog nodes can communicate with each other through wired or real-time and low-latency services is presented. It is shown that the mean
wireless transmission. Moreover, fog nodes have some general energy expenditure in fog computing is 40.48% less than that in a con-
computing capabilities. In particular, those fog nodes engaged in ventional cloud computing model. Evaluation results show that fog
enhanced analytics need to configure accelerator modules such as computing is an improved, eco-friendly computing platform that can
graphics processing units, field programmable gate arrays, and digital support IoT better compared to the existing cloud computing paradigm
signal processors to provide supplementary computational throughput. [64]. To take advantage of edge computing and to complement central-
Many types of storage are required in fog nodes to meet the required ized cloud computing, a portion of IoT applications that are
reliability and data integrity of the system and scenario. Generally, there energy-efficient in fog computing architecture should be identified. In
are a rich set of sensors and actuators at the edge of the network in an order to compare the energy consumption of applications using central-
application scenario. These sensors and actuators are connected to the ized DCs in cloud computing with applications using nano Data Centers
fog node via a multitude of interfaces, such as PCIe, USB, and Ethernet. (nDCs) based on fog computing, flow-based and time-based energy
Fog nodes can be worked in a mesh manner to provide load balancing, consumption models for shared and unshared network equipment are
resilience, fault tolerance, data sharing, and minimization of cloud proposed in Ref. [65]. Correspondingly, a set of measurements and ex-
communication. There are often three tiers in a fog computing system, periments are used to provide data for the models, in which nano-servers
but more tiers can be allowed for a special application scenario. At the in fog computing are implemented.
edge of the network, fog nodes are typically focused on sensor data The results indicate that the best energy savings using nDCs can be
acquisition/collection, data normalization, and command/control of attained for some applications that generate and distribute a large
sensors and actuators. At the next higher tier, fog nodes are focused on amount of data in end-user premises with low access data rate, such as
data filtering, compression, and transformation. At the higher tiers or video surveillance in end-users’ homes. The tradeoff between power
nearest the backend cloud, fog nodes are focused on aggregating data and consumption and transmission delay in the fog-cloud computing system
turning the data into knowledge. Architecturally, fog nodes at the edge is investigated in Ref. [66]. The segmentation of what tasks go to the edge
may require less processing, communications, and storage than nodes at and what tasks go to the backend cloud is application specific. Simulation
high levels. However, Input and Output (I/O) accelerators required to results on the user case of medical emergency service demonstrate the
facilitate sensor data intake at the edge are much larger in aggregate than benefits of coordinated control and management of a combined fog and
I/O accelerators designed for higher-level nodes. With the increase in the cloud system. Thus, the design of a coordinated management strategy
number of tiers, each tier would be sifting and extracting meaningful data becomes critical and needs to address the different cloud/edge resources
to create more intelligence. in a joint framework capable of managing the emerging edge-to-cloud
The conventional centralized cloud computing continues to remain an computing and network architecture [67].
important part of computing systems as fog computing emerges. Cloud
and edge computing complement each other to form a mutually benefi-
4.4. Fog-RAN: fog computing in radio access networks
cial and inter-dependent service continuum. Some functions are naturally
more advantageous to carry out in centralized cloud, while others are
The 5G RAN seamlessly and ubiquitously connects everything, brings
better suited at the edge. In Ref. [63], a quantitative analysis of energy
a 1000-fold increase in terms of area capacity, supports 100 billion
consumption in a scenario where 25% of the IoT applications demand
connected wireless devices, and provides diversified use cases as well as

83
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

high QoS requirements of multimedia applications, compared with cur- Access Point (F-AP), and the edge caching in F-APs is a key technology to
rent 4G LTE networks [68]. To achieve the above goals, C-RAN has been improve SE and EE in maintaining a low latency level [73]. In addition,
proposed as a combination of emerging technologies from both the the MEC server is compatible with the conventional wireless network
wireless network and cloud computing [69]. The conventional BS is architecture, which can be deployed at multiple locations, such as at the
decoupled into two parts: the baseband resources are pooled at BaseBand eNodeB, RNC, and RAT cell aggregation sites. The system architecture of
Units (BBUs) in a centralized location, remote radio heads with radio F-RANs is evolved from HetNets and C-RANs, and F-AP is introduced to
frequency functions connect with the BBU pool through the wire/- integrate not only the front Radio Frequency (RF), but also the limited
wireless fronthaul links. Based on centralized cloud principle of sharing caching, the locally distributed CRSP, and simple CRRM functional
storage and computing resources via virtualization, Cloud RANs capabilities.
(C-RANs) bring the advantages of high spectral efficiency and energy
efficiency while at the same time reducing the cost of network deploy- 5. Open issues and challenges
ment and operation. However, the practical fronthaul is often capacity
constrained or time-delay constrained, which presents a bottleneck to the Table 3 summarizes the main features of these three edge computing
capacity of C-RANs [70]. To overcome the disadvantages of C-RANs with technologies. Fog computing is initiated to address some challenges in
fronthaul constraints, heterogeneous C-RANs (H-CRANs) are proposed as meeting new requirements of IoTs. MEC is recognized as one of the key
a potential solution [71]. Compared to the C-RAN architecture, the technologies to meet 5G requirements. Cloudlets are proposed to address
proposed H-CRAN alleviates the fronthaul requirements with the some challenges in mobile computing. From the application point of
participation of High-Power Nodes (HPNs) (e.g., macro or micro base view, MEC enables an open RAN, which can host third party innovative
station). The user and control planes are decoupled in H-CRANs. HPNs applications and content at the edge of the network. Cloudlet enables
execute the functions of the control plane, which delivers all control new classes of mobile applications that are both compute-intensive and
signaling and system broadcasting data to UEs. HPNs are connected to latency-sensitive in an open ecosystem based on cloudlets. Fog
the BBU pool via the backhaul links, which alleviates the constraints on computing enables high-performance, interoperability, and security in a
fronthaul. However, H-CRANs still have the same challenges in practice. multi-vendor fog computing-based ecosystem. The similarity between
More and more IoT applications generate unprecedented amount of data the three technologies is openness. Operators open their networks to
at the edge of networks, which worsens the fronthaul constraints. In third parties to deploy innovative applications and services.
addition, processing and storage capabilities in edge devices have not In this section, some of the pertinent open issues that require addi-
been fully exploited; this is a promising approach to successfully alleviate tional investigations for edge computing are summarized.
the burden of the fronthaul and BBU pool. Taking full advantage of fog
computing and C-RANs, Fog Radio Access Networks (F-RANs) have been 5.1. Big data mining in edge computing
proposed to tackle these aforementioned disadvantages of C-RANs and
H-CRANs [72]. In F-RANs, local radio signal processing, cooperative To adapt the massive types of packet traffic and the time-varied radio
radio resource management, and distributed storage capabilities in edge channel, edge computing should be information-aware and reconfigured.
devices can decrease the heavy burden on the fronthaul and avoid Big data, described by volume, variety, velocity, and value, include
large-scale radio signal processing in the centralized BBU pool. There- subscriber-level, cell-level, core-network-level, and other level data, can
fore, F-RANs can achieve high SE/EE, low latency, and excellent reli- facilitate the network toward a more proactive one [74]. Owing to the
ability to meet 5G requirements. fast development of big data mining, it is feasible to utilize big data
Table 2 presents the differences between MEC and F-RANs. First, F- technology to extract interesting patterns or knowledge to enhance the
RANs have been proposed as an enhancement and evolution of C-RANs to self-organizing capabilities in edge computing. Hierarchical data mining
overcome the disadvantages of C-RANs with the fronthaul constraints. techniques should be used. The transmission of large volume of data
MEC is based on a virtualized platform to enable an open RAN, which can collected by edge devices puts a heavy burden on the fronthaul/back-
host third party innovative applications and content at the edge of the haul, and as a result, data mining can be pre-executed in the edge devices.
network. Operators can open their networks to authorized third parties, In the centralized cloud computing, this pre-executed information is
exposing capabilities to over the top players and application developers reprocessed. Meanwhile, the computing of sparse, uncertain and
to flexibly and rapidly deploy innovative applications and services to- incomplete data is a big problem, which requires advanced data mining
ward mobile subscribers, enterprises, and vertical segments. MEC does algorithms [75].
not contradict with C-RANs but rather complement them. MEC is mainly
compute-oriented, which enables running computation-intensive tasks 5.2. Network slicing in edge computing
for edge users by offloading energy-consuming computations of the ap-
plications to the MEC server to enhance user experience. In F-RANs, the To meet the diverse use cases and business models for the emerging
Cooperative Radio Signal Processing (CRSP) and Cooperative Radio applications of mobile internet and IoT, both revolutionary wireless
Resource Management (CRRM) can be executed in Fog-computing-based network architectures and advanced technologies are anticipated. As a
result, network slicing is proposed recently to flexibly provide SDN in a
cost-efficient way. In the concept of network slicing [76], the network
Table 2 entity is sliced into multiple isolated network slice instances, and each
Differences between MEC and F-RANs. slice instance has appropriate network functions and uses advanced radio
MEC F-RANs access technologies for a specific use case or business model. By exploring
Motivation Enable an open radio access Overcome the disadvantages of
SDN and NFV, network slice instances and the isolation between them
network that can host third the fronthaul constraints with can be conveniently realized [77]. Despite the evident attractive ad-
party innovative applications limited capacity and long delay vantages in centralized cloud computing, network slicing comes with
and content at the edge of the several severe challenges when applied in edge computing. First, the
network
conventional creation of network slice instance is mainly business driven.
Relationship Complement with C-RANs An enhancement and evolution
with C-RANs of C-RANs The network slicing solution mainly addresses the requirements of
Key technology Computation offloading Edge caching different services, which do not highlight the characteristics of edge
Deployment Be compatible with A new system architecture is computing on network slicing creation. For example, when the radio
scheme conventional wireless network evolved from HetNets and C- resource in RANs is in shortage, the requested network slicing may not be
architecture RANs by introducing F-AP
effective. As a result, the network slicing should consider radio

84
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

Table 3
Comparison of cloudlets, MEC, and fog computing.
Item Mobile edge computing Cloudlets Fog computing

Organization ETSI MEC supported by Huawei, IBM, OEC launched by Vodafone, Intel, Huawei, and OpenFog Consortium founded by ARM, Cisco, Dell, Intel,
Intel, Nokia Networks, NTT DoCoMo, and Carnegie Mellon University Microsoft, and Princeton University
Vodafone
Is the corresponding Yes No (influences other standards organizations No (OpenFog has an affiliation agreement with IEEE and
consortium a such as ETSI MEC) will be establishing liaisons with other standards
standards body? organizations, with the objective of collaborating in the
creation of standards)
Which business 5G requirements in the Some applications based on mobile computing Internet of Things
interests are driving? telecommunications industry
Motivation from the enable an open RAN that can host third enable new classes of mobile applications that Enable high-performance, interoperability and security in
application point of party innovative applications and content are both compute-intensive and latency- a multi-vendor fog computing-based ecosystem
view at the edge of the network sensitive in an open ecosystem based on
cloudlets
Features on openness Operators open their networks to third OPENSTACKþþ is an open source platform The OpenFog Reference Architecture is used as a common
parties to deploy innovative applications that extends the functionality of Openstack to baseline for achieving a multivendor interoperable fog
and services support cloudlets computing ecosystem

transmission impacts, and the corresponding network slicing jointly [6] M.A. Razzaque, M. Milojevic-Jevric, A. Palade, S. Cla, Middleware for internet of
things: a survey, IEEE Internet Things J. 3 (2016) 70–95.
considering the status of RANs should be defined. Second, most of the
[7] A. Brogi, S. Forti, QoS-aware deployment of IoT applications through the fog, IEEE
existing work on network slicing is purely based on CNs, while network Internet Things J. (2017), https://doi.org/10.1109/JIOT.2017.2701408.
slicing as an end-to-end solution should cover the specific characteristics [8] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, W. Zhao, A survey on internet of things:
of RANs. To overcome these challenges, a framework solution for a new architecture, enabling Technologies, security and privacy, and applications, IEEE
Internet Things J. (2017).
network slicing technique on edge computing is anticipated. [9] M. Satyanarayanan, The emergence of edge computing, Computer (Long. Beach.
Calif) 50 (2017) 30–39.
6. Conclusion [10] N. Bizanis, F. Kuipers, SDN and virtualization solutions for the Internet of Things: a
survey, IEEE Access. J. 4 (2016) 5591–5606.
[11] J. Li, M. Peng, A. Cheng, Y. Yu, C. Wang, Resource allocation optimization for
This paper outlines and surveys the state-of-the-art edge computing delay-sensitive traffic in fronthaul constrained cloud radio access networks, IEEE
technologies. With the goal of understanding further intricacies of the Syst. J. (2014) 1–12.
[12] ETSI, Mobile-edge Computing Introductory Technical White Paper, White Paper,
key technologies, we have broadly divided the body of knowledge into Mobile-edge Computing Industry Initiative, 2014, https://portal.etsi.org/portals/0/
cloudlet, mobile edge computing, and fog computing. Within each of tbpages/mec/docs/mobile-edge_computing_– _introductory_technical_white_paper_
these aspects, we have given a detailed tutorial on the principle, system v1.
[13] Y. Mao, C. You, J. Zhang, K. Huang, K.B. Letaief, A Survey on Mobile Edge
architecture, standards, and applications. Nevertheless, given the relative Computing: the Communication Perspective, 2017. https://arxiv.org/abs/1701.
infancy of the field, there are still many outstanding problems that 01090.
require further investigation from the perspective of key techniques and [14] Guenter I. Klas, Fog Computing and Mobile Edge Cloud Gain Momentum Open Fog
Consortium, ETSI MEC and Cloudlets, 2015. http://yucianga.info/?p¼938.
advanced solutions. Given the extensiveness of the research areas, it is
[15] M. Satyanarayanan, P. Bahl, R. Caceres, N. Davies, The case for VM-base cloudlets
also concluded that more rigorous investigations are required with in mobile computing, Pervasive Comput. 8 (2009) 14–23.
greater attention to be focused on transforming well-established fog [16] Nokia Solutions and Networks, Increasing Mobile Operators' Value Proposition with
computing into fog computing-based RANs. Furthermore, with the Edge Computing, White Pap, 2013, pp. 1–6, http://nsn.com/portfolio/liquid-net/
intelligent-broadband-management/liquid-applications.
introduction of the advanced big data mining and network slicing, the [17] Y. Ch. Hu, M. Patel, D. Sabella, N. Sprecher, V. Young, Mobile Edge Computing a
availability of varied degrees of freedom along with the associated con- Key Technology towards 5G, 2015. http://10.3.200.202/cache/8/03/etsi.org/
straints further beckon the design and validation of the original models in 6e14a9668574b8b935111768d9f6e501/etsi_wp11_mec_a_key_technology_
towards_5g.pdf.
the context of edge computing. [18] N. Sprecher, J. Friis, R. Dolby, J. Reister, Edge Computing Prepares for a Multi-
access Future, 2016. http://www.telecomtv.com/articles/mec/edge-computing-
Acknowledgments prepares-for-a-multi-access-future-13986/.
[19] F. Bonomi, R. Milito, J. Zhu, S. Addepalli, Fog Computing and its Role in the
Internet of Things, Proc. First Ed. MCC Work. Mob. Cloud Comput., 2012,
This work was supported in part by the National Natural Science pp. 13–16.
Foundation of China under Grant No. 61361166005, the National High [20] Z. Sanaei, S. Abolfazli, A. Gani, R. Buyya, Heterogeneity in mobile cloud computing:
taxonomy and open challenges, IEEE Commun. Surv. Tutor. 16 (2014) 369–392.
Technology Research and Development Program of China under Grant [21] K. Ha, P. Pillai, W. Richter, Y. Abe, M. Satyanarayanan, Just-in-time provisioning
No. 2014AA01A701, the National Basic Research Program of China (973 for cyber foraging, in: MobiSys 2013-Proc. 11th Annu. Int. Conf. Mob. Syst. Appl.
Program) under Grant No. 2013CB336600, and the State Major Science Serv, 2013, pp. 153–166.
[22] Y. Gao, W. Hu, K. Ha, B. Amos, et al., Are Cloudlets Necessary? School of Computer
and Technology Special Projects (Grant No. 2016ZX03001020-006).
Sci. Carnegie Mellon Univ., Pittsburgh, PA, USA, 2015. Tech. Rep. CMU-CS-15-139,
http://reports-archive.adm.cs.cmu.edu/anon/anon/usr/ftp/usr0/ftp/2015/CMU-
References CS-15-139.pdf.
[23] K. Ha, Y. Abe, Z. Chen, W. Hu, et al., Adaptive VM Handoff across Cloudlets,
[1] M. Peng, K. Zhang, Recent advances in fog radio access networks: performance Technical Report CMU-CS-15-113, CMU School of Computer Science, 2015, http://
analysis and radio resource allocation, IEEE Access J. 4 (2016) 5003–5009. ra.adm.cs.cmu.edu/anon/2015/CMU-CS-15-113.pdf.
[2] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, M. Ayyash, Internet of [24] M. Satyanarayanan, R. Schuster, M. Ebling, G. Fettweis, H. Flinck, K. Joshi,
things: a survey on enabling technologies, protocols, and applications, IEEE K. Sabnani, An open ecosystem for mobile-cloud convergence, IEEE Commun. Mag.
Commun. Surv. Tutor. 17 (2015) 2347–2376. 53 (2015) 63–70.
[3] CISCO, The Internet of Things How the Next Evolution of the Internet Is Changing [25] K. Ha, Z. Chen, W. Hu, W. Richter, P. Pillai, M. Satyanarayanan, Towards wearable
Everything, White Pap, 2011, http://www.cisco.com/c/dam/en us/about/ac79/ cognitive assistance, in: Proc. 12th Annu. Int. Conf. Mob. Syst. Appl. Serv. -
docs/innov/IoT_IBSG_0411FINAL.pdf. MobiSys'14, 2014, pp. 68–81.
[4] M. Chiang, T. Zhang, Fog and IoT: an overview of research opportunities, IEEE [26] M. Satyanarayanan, P. Simoens, Y. Xiao, P. Pillai, Z. Chen, K. Ha, W. Hu, B. Amos,
Internet Things J. 3 (2016) 854–864. Edge analytics in the internet of things, IEEE Pervasive Comput. 14 (2015) 24–31.
[5] F. Ganz, D. Puschmann, P. Barnaghi, F. Carrez, A practical evaluation of information [27] M. Satyanarayanan, G. Lewis, E. Morris, S. Simanta, J. Boleng, K. Ha, The role of
processing and abstraction techniques for the internet of things, IEEE Internet cloudlets in hostile environments, IEEE Pervasive Comput. 12 (2013) 40–49.
Things J. 2 (2015) 340–354.

85
Y. Ai et al. Digital Communications and Networks 4 (2018) 77–86

[28] K. Ha, M. Satyanarayanan, OpenStackþþ for Cloudlet Deployment, School of [52] Open Fog Consortium, [Online]. https://www.openfogconsortium.org/.
Computer Science Carnegie Mellon University Pittsburgh, 2015. http://elijah.cs. [53] OpenFog Consortium Architecture Working Group, OpenFog Architecture Overview
cmu.edu/DOCS/CMU-CS-15-123.pdf. White Paper. https://www.openfogconsortium.org/wp-content/uploads/OpenFog-
[29] ETSI GS MEC 001, Mobile Edge Computing (MEC) Terminology V1.1.1, 2016. Architecture-Overview-WP-2-2016.pdf.
http://www.etsi.org/deliver/etsi_gs/MEC/001_099/001/01.01.01_60/gs_ [54] Open Fog Consortium, OpenFog Reference Architecture for Fog Computing.
MEC001v010101p.pdf. https://www.openfogconsortium.org/wp-content/uploads/OpenFog_Reference_
[30] ETSI GS MEC 002, Mobile Edge Computing (MEC) Technical Requirements V1.1.1, Architecture_2_09_17-FINAL-1.pdf.
2016. http://www.etsi.org/deliver/etsi_gs/MEC/001_099/002/01.01.01_60/gs_ [55] A.V. Dastjerdi, R. Buyya, Fog computing: helping the Internet of Things realize its
MEC002v010101p.pdf. potential, Computer 49 (2016) 112–116.
[31] ETSI GS MEC 003, Mobile Edge Computing (MEC) Framework and Reference [56] S.K. Sharma, X. Wang, Live data analytics with collaborative edge and cloud
Architecture V1.1.1, 2016. http://www.etsi.org/deliver/etsi_gs/MEC/001_099/ processing in wireless IoT networks, IEEE Access 5 (2017) 4621–4635.
003/01.01.01_60/gs_MEC003v010101p.pdf. [57] B. Tang, Z. Chen, G. Hefferman, S. Pei, W. Tao, H. He, Q. Yang, Incorporating
[32] ETSI GS MEC-IEG 004, Mobile Edge Computing (MEC) Service Scenarios V1.1.1, intelligence in fog computing for big data analysis in smart cities, IEEE Trans. Ind.
2015. http://www.etsi.org/deliver/etsi_gs/MEC/001_099/004/01.01.01_60/gs_ Inf. 13 (2017) 2140–2150, https://doi.org/10.1109/TII.2017.2679740.
MEC003v010101p.pdf. [58] S. Basudan, X. Lin, K. Sankaranarayanan, A privacy-preserving vehicular
[33] ETSI GS MEC-IEG 005, Mobile Edge Computing (MEC) Proof of Concept Framework crowdsensing based road surface condition monitoring system using fog computing,
V1.1.1, 2015. http://www.etsi.org/deliver/etsi_gs/MEC-IEG/001_099/005/01.01. IEEE Internet Things J. 4 (2017) 772–782.
01_60/gs_MEC-IEG005v010101p.pdf. [59] P. Hu, H. Ning, T. Qiu, H. Song, Y. Wang, X. Yao, Security and privacy preservation
[34] ETSI GS MEC-IEG 006, Mobile Edge Computing Market Acceleration MEC Metrics scheme of face identification and resolution framework using fog computing in
Best Practice and Guidelines V1.1.1, 2017. http://www.etsi.org/deliver/etsi_gs/ internet of things, IEEE Internet Things J. (2017), 1–1.
MEC-IEG/001_099/006/01.01.01_60/gs_MEC-IEG006v010101p.pdf. [60] X. Hou, Y. Li, M. Chen, D. Wu, D. Jin, S. Chen, Vehicular fog computing: a viewpoint
[35] T. Taleb, S. Dutta, A. Ksentini, M. Iqbal, H. Flinck, Mobile edge computing potential of vehicles as the infrastructures, IEEE Trans. Veh. Technol. 65 (2016) 3860–3873.
in making cities smarter, IEEE Commun. Mag. 55 (2017) 38–43. [61] M. Aazam, E.N. Huh, Fog computing: the cloud-IoT/IoE middleware paradigm, IEEE
[36] S. Lin, H.F. Cheng, W. Li, Z. Huang, P. Hui, C. Peylo, Ubii: physical world interaction Potentials 35 (2016) 40–44.
through augmented reality, IEEE Trans. Mob. Comput. 16 (2017) 872–885. [62] L. Gu, D. Zeng, S. Guo, A. Barnawi, Y. Xiang, Cost efficient resource management in
[37] N. Kumar, S. Zeadally, J.J.P.C. Rodrigues, Vehicular delay-tolerant networks for fog computing supported medical cyber-physical system, IEEE Trans. Emerg. Top.
smart grid data management using mobile edge computing, IEEE Commun. Mag. 54 Comput. 5 (2017) 108–119.
(2016) 60–66. [63] S. Misra, S. Sarkar, Theoretical modelling of fog computing: a green computing
[38] X. Sun, N. Ansari, EdgeIoT: mobile edge computing for the Internet of Things, IEEE paradigm to support IoT applications, IET Netw. 5 (2016) 23–29.
Commun. Mag. 54 (2016) 22–29. [64] S. Sarkar, S. Chatterjee, S. Misra, Assessment of the suitability of fog computing in
[39] A. Ceselli, M. Premoli, S. Secci, Mobile edge cloud network design optimization, the context of internet of things, IEEE Trans. Cloud Comput. (2015).
IEEE/ACM Trans. Netw. 25 (2017) 1818–1831. [65] F. Jalali, K. Hinton, R. Ayre, T. Alpcan, R.S. Tucker, Fog computing may help to save
[40] S. Barbarossa, S. Sardellitti, P. Di Lorenzo, Communicating while computing: energy in cloud computing, IEEE J. Sel. Areas Commun. 34 (2016) 1728–1739.
distributed mobile cloud computing over 5G heterogeneous networks, IEEE Signal [66] R. Deng, R. Lu, C. Lai, T.H. Luan, H. Liang, Optimal workload allocation in fog-cloud
Process. Mag. 31 (2014) 45–55. computing toward balanced delay and power consumption, IEEE Internet Things J.
[41] P. Mach, Z. Becvar, Mobile Edge Computing: a survey on architecture and (2016) 1171–1181.
computation offloading, IEEE Commun. Surv. Tutor. 19 (2017) 1628–1656, https:// [67] X. Masip-Bruin, E. Marín-Tordera, G. Tashakor, A. Jukan, G.J. Ren, Foggy clouds
doi.org/10.1109/COMST.2017.2682318. and cloudy fogs: a real need for coordinated management of fog-to-cloud computing
[42] H. Flores, P. Hui, S. Tarkoma, Y. Li, S. Srirama, R. Buyya, Mobile code offloading: systems, IEEE Wirel. Commun. 23 (2016) 120–128.
from concept to practice and beyond, IEEE Commun. Mag. 53 (2015) 80–88. [68] M. Peng, Y. Li, Z. Zhao, C. Wang, System architecture and key technologies for 5G
[43] K. Zhang, Y. Mao, S. Leng, Q. Zhao, L. Li, X. Peng, L. Pan, S. Maharjan, Y. Zhang, heterogeneous cloud radio access networks, IEEE Netw. 29 (2015) 6–14.
Energy-efficient offloading for mobile edge computing in 5G heterogeneous [69] M. Peng, Y. Sun, X. Li, Z. Mao, C. Wang, Recent advances in cloud radio access
networks, IEEE Access. J. 4 (2016) 5896–5907. networks: system architectures, key techniques, and open issues, IEEE Commun.
[44] X. Chen, L. Jiao, W. Li, X. Fu, Efficient multi-user computation offloading for Surv. Tutor. 18 (2016) 2282–2308.
mobile-edge cloud computing, IEEE/ACM Trans. Netw. 24 (2016) 2795–2808. [70] M. Peng, C. Wang, V. Lau, H.V. Poor, Fronthaul-constrained cloud radio access
[45] C. You, K. Huang, H. Chae, B.H. Kim, Energy-efficient resource allocation for networks: insights and challenges, IEEE Wirel. Commun. 22 (2015) 152–160.
mobile-edge computation offloading, IEEE Trans. Wirel. Commun. 16 (2017) [71] M. Peng, Y. Li, J. Jiang, J. Li, C. Wang, Heterogeneous cloud radio access networks:
1397–1411. a new perspective for enhancing spectral and energy efficiencies, IEEE Wirel.
[46] S. Sardellitti, G. Scutari, S. Barbarossa, Joint optimization of radio and Commun. 21 (2014) 126–135.
computational resources for multicell mobile-edge computing, IEEE Trans. Signal [72] M. Peng, S. Yan, K. Zhang, C. Wang, Fog-computing-based radio access networks:
Inf. Process. Over Netw. 1 (2015) 89–103. issues and challenges, IEEE Netw. 30 (2016) 46–53.
[47] S. Secci, P. Raad, P. Gallard, Linking virtual machine mobility to user mobility, IEEE [73] S. Jia, Y. Ai, Z. Zhao, M. Peng, C. Hu, Hierarchical content caching in fog radio
Trans. Netw. Serv. Manag. 13 (2016) 927–940. access networks: ergodic rate and transmit latency, China Commun. 13 (2016)
[48] P. Raad, S. Secci, D.C. Phung, A. Cianfrani, P. Gallard, G. Pujolle, Achieving sub- 1–14.
second downtimes in large-scale virtual machine migrations with LISP, IEEE Trans. [74] A. Imran, A. Zoha, Challenges in 5G: how to empower SON with big data for
Netw. Serv. Manag. 11 (2014) 133–143. enabling 5G, IEEE Netw. 28 (2014) 27–33.
[49] T. Taleb, A. Ksentini, P. Frangoudis, Follow-me cloud: when cloud services follow [75] X. Wu, X. Zhu, G.Q. Wu, W. Ding, Data mining with big data, IEEE Trans. Knowl.
mobile users, IEEE Trans. Cloud Comput. (2016), 1–1. Data Eng. 26 (2014) 97–107.
[50] X. Sun, N. Ansari, PRIMAL: PRofIt Maximization Avatar pLacement for mobile edge [76] R. Hattachi, J. Erfanian, 5G White Paper, NGMN Alliance, 2015.
computing, in: 2016 IEEE Int. Conf. Commun. ICC 2016, 2016, pp. 1–6. [77] Q. Li, G. Wu, A. Papathanassiou, U. Mukherjee, An End-to-end Network Slicing
[51] J. Plachy, Z. Becvar, E.C. Strinati, Dynamic resource allocation exploiting mobility Framework for 5G Wireless Communication Systems, 2016. https://arxiv.org/abs/
prediction in mobile edge computing, in: IEEE Int. Symp. Pers. Indoor Mob. Radio 1608.00572.
Commun. PIMRC, 2016.

86

You might also like