Integration of Networking, Caching, and Computing in Wireless Systems: A Survey, Some Research Issues, and Challenges

IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO.
1, FIRST QUARTER 2018 7
Integration of Networking, Caching, and Computing

in Wireless Systems: A Survey, Some
Research Issues, and Challenges
Chenmeng Wang, Ying He, Student Member, IEEE, F. Richard Yu , Senior Member, IEEE,
Qianbin Chen, and Lun Tang
Abstract—Since the recently emerging mobile applications have wireless networks solely focusing on communication are no
posed significant demands not only on high data rate but also longer capable of meeting the demand raised by such appli-
on high caching and computing capabilities, the growth in com- cations not only on high data rate, but also on high caching
munication capability alone is no longer sustainable for wireless
networks. The integration of networking, caching, and computing and computing capabilities [2]. Although the pivotal role of
functionalities into one system can provide not only native sup- communication in wireless networks can never be overempha-
port for highly scalable and efficient content retrieval, but also sized, the growth in communication alone is not sustainable
powerful capability of data processing, hence reducing duplicate any longer [3]. On the other hand, recent advances in commu-
content transmissions and enabling swift executions of com- nication and information technologies have fueled a plethora
putationally intensive tasks. Despite the prospect of integrated
networking, caching, and computing systems, a number of sig- of innovations in various areas, including networking, caching
nificant research challenges remain to be addressed prior to and computing, which have the potential to profoundly impact
widespread deployment of integrated networking, caching, and our society via the developments of smart homes, smart
computing systems, including latency requirement, interfaces, transportation, smart cities [4], etc.
mobility management, resource and architecture tradeoffs, con- Therefore, the integration of networking, caching and com-
vergence, etc. In this paper, we provide a brief survey on some
of the works that have been done to enable the integrated puting into one system becomes a natural trend [5], [6].
networking, caching, and computing system, and discuss sev- By incorporating caching functionality into the network, the
eral research challenges. We identify a number of important system can provide native support for highly scalable and effi-
aspects of the integration of networking, caching, and com- cient content retrieval, and meanwhile, the duplicate content
puting: motivations, frameworks, performance metrics, enabling transmissions within the network can be reduced significantly.
technologies, and challenges. At last, some broader perspectives
are explored. As a consequence, mobility, flexibility and security of the
network can be considerably improved [7], [8]. On the other
Index Terms—Networking, caching, computing, wireless. hand, the incorporation of computing functionality endows the
network with powerful capability of data processing, hence
enabling the execution of computationally intensive appli-
I. I NTRODUCTION cations within the network. By offloading mobile devices’
computation tasks (entirely or partially) to resource-rich com-
EW MOBILE applications, such as natural language
N processing, augmented reality and face recognition [1],
are emerging rapidly in recent years. However, conventional
puting infrastructures in vicinity or in remote clouds, the
task execution time can be considerably reduced, and the
local resources of mobile devices especially battery life can
Manuscript received March 16, 2017; revised August 10, 2017; accepted be conserved, thus enriching the computing experience of
September 16, 2017. Date of publication October 6, 2017; date of cur- mobile users [9]. Moreover, networking, caching and comput-
rent version February 26, 2018. This work was supported in part by
the Graduates Research Innovation Program of Chongqing under Grant ing functionalities can complement and reinforce each other by
CYB15106, in part by the National Natural Science Foundation of China interactions. For instance, some of the computation results can
under Grant 61571073, in part by the National High Technology Research be cached for future use, thus alleviating backhaul workload.
and Development Program of China under Grant 2014AA01A701, and in
part by the Natural Sciences and Engineering Research Council of Canada. On the other hand, some cached content can be transcoded
(Corresponding author: F. Richard Yu.) into other versions to better suit specific user demands, thus
C. Wang, Q. Chen, and L. Tang are with the Chongqing Key Laboratory economizing storage spaces and maximizing the utilization of
of Mobile Communications Technology, Chongqing University of Posts
and Telecommunications, Chongqing 400065, China (e-mail: wangchen- caching.
[email protected]; [email protected]; [email protected]). Despite the potential vision of integrated networking,
Y. He is with the School of Information and Communication Engineering, caching and computing systems, a number of significant
Dalian University of Technology, Dalian 116024, China, and also with the
Department of Systems and Computer Engineering, Carleton University, research challenges remain to be addressed before widespread
Ottawa, ON K1S 5B6, Canada (e-mail: [email protected]). application of the integrated systems, including latency
F. R. Yu is with the Department of Systems and Computer requirement, bandwidth constraints, interfaces, mobility man-
Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada (e-mail:
[email protected]). agement, resource and architecture tradeoffs, convergence as
Digital Object Identifier 10.1109/COMST.2017.2758763 well as non-technical issues such as governance regulations,
1553-877X c 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
8 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 20, NO. 1, FIRST QUARTER 2018
Fig. 1. Road map of the integration of networking, caching and computing systems.
etc. Particularly, the resource allocation issue, in which the Section V presents some performance metrics that are neces-
management and allocation of three types of resources should sary to evaluate the quality and performance of the integrated
be jointly considered to effectively and efficiently serve the systems. Some of the enabling technologies are summarized
user requirements, is especially challenging. In addition, non- in Section VI according to different types of integrations. A
trivial security challenges are induced by a large number of number of research challenges and broader perspectives are
intelligent devices/nodes with self adaptation/context aware- discussed in Section VII. Finally, we conclude this article in
ness capabilities in the integrated system. These challenges Section VIII.
need to be broadly tackled through comprehensive research
efforts. II. A N OVERVIEW OF N ETWORKING , C ACHING
In this article, we provide a brief survey on some of AND C OMPUTING IN W IRELESS S YSTEMS
the works that have already been done to enable the inte-
In this section, we present a brief overview on the history
gration of networking, caching and computing systems, and
and current research progresses of networking, caching and
explore several research challenges. A taxonomy graph of
computing technologies in wireless systems.
our approach towards the design of integrated networking,
caching and computing system is presented as Fig. 1. As
shown in the figure, we identify seven aspects of the inte- A. Recent Advances in Wireless Networking
gration of networking, caching and computing systems, on The communication and networking technologies of the
which we would like to focus: overview, motivations, frame- fifth generation (5G) wireless telecommunication networks
works, performance metrics, enabling technologies, challenges are being standardized at full speed along the time line
and broader perspectives. of International Mobile Telecommunications for 2020 (IMT-
The rest of the article is organized as follows. Section II 2020), which is proposed by International Telecommunication
presents an overview on the history and current research Union (ITU) [10]. Consequently, the unanimous agreement
progresses of wireless networking, caching and computing on the prospect of a new, backward-incompatible radio access
technologies. Section III discusses the motivations and require- technology is emerging in industry.
ments of the integration of networking, caching and comput- Significant improvements on data rate and latency
ing. In Section IV we explore some of the typical frameworks performance will be powered by logical network
of the integrated networking, caching and computing systems. slices [11], [12] in 5G communication environment.
WANG et al.: INTEGRATION OF NETWORKING, CACHING, AND COMPUTING IN WIRELESS SYSTEMS 9
Furthermore, network flexibility will be greatly promoted project METIS in its final deliverable [35]. In massive MIMO
by dynamic network slices, therefore providing possibility scenario, K user equipments (UEs) are served on the same
to the emergence of a variety of new network services time-frequency resource by a base station (BS) equipped with
and applications. Due to the fact that the key problems of M antennas, where M K. The deployment of a large
network slices are how to properly slice the network and number of antennas at the BS leads to a particular prop-
at what granularity [11], it is not difficult to predict that a agation scenario named favorable propagation, where the
number of already existing technologies, such as software wireless channel becomes near-deterministic since the BS-to-
defined networking (SDN) [13] and network functions UE radio links become near-orthogonal to each other [36].
virtualization (NFV) [14], will be taken into consideration. Recent advances in fields like 3D MIMO, hybrid beamform-
SDN can enable the separation between data plane and con- ing (BF) and understanding of the asymptotic behavior of
trol plane, therefore realizing the independence of the data random matrices suggest that massive MIMO has the potential
plane capacity from the control plane resource, which means to bring unprecedented gains in terms of spectrum and energy
high data rate can be achieved without incurring overhead efficiencies and robustness to hardware failures [35].
upon control plane [15]–[17]. Meanwhile, the separation of With the severe shortage in currently utilized spectrum
data and control plane endows high programmability, low- (sub-6 GHz), the utilization of higher frequency bands, e.g.,
complexity management, convenient configuration and easy millimeter-wave (mmWave), becomes a consensus of the
troubleshooting to the network [18]. The controller in control communication community [37]. Working on 3–300 GHz
plane and the devices in data plane are bridged by a well- bands, mmWave communications assisted by smart antenna
defined application programming interface (API), of which arrays can track and transmit signals to high-speed targets
a well known example is OpenFlow [19]. The instructions over a long distance [38]. Since the physical antenna array
of a controller to the devices in data plane are transmitted size can be greatly reduced due to decrease in wavelength,
through flow tables, each of which defines a subset of the mmWave communications are appealing to large-scale antenna
traffic and the corresponding action. Due to the advantages systems [39]. On the other hand, the detrimental effects of
described above, SDN is considered promising on providing the high propagation loss in mmWave communications can
programmable network control both in wireless and wired be neatly compensated by the achievable high beamforming
networks [20], [21]. gains with massive MIMO [37], [40]. Therefore, the mmWave
While NFV presents the technology that decouples the communication assisted by massive MIMO is envisaged as a
services from the network infrastructures that provide them, promising solution for future telecommunications.
therefore maximizing the utilization of network infrastructure Another way to deal with the spectrum scarcity is to increase
by offering the possibility that services from different service the utilization efficiency of spectrum. Allowing two UEs
providers can share the same network infrastructure [14], [22]. to perform direct data transmissions, device-to-device (D2D)
Furthermore, easy migration of new technology in cohabitation communications enable a flexible reuse of radio resources,
with legacy technologies in the same network can be realized and therefore improve the spectral efficiency and ease core
by isolating part of the network infrastructure [14]. network data processing workloads [41], [42]. Moreover,
On the other hand, heterogeneous network (HetNet) has the combination of D2D communications with information-
been recently proposed as a promising network architecture centric networking (ICN) is worth considering. Despite the
to improve network coverage and link capacity [23], [24]. small-sized storage of UEs, the ubiquitous caching capabil-
By employing different sizes of small cells and various radio ity residing in UEs should not be neglected, due to the UEs’
access technologies in one macro cell, energy efficiency (EE) pervasive distribution and ever-increasing storage sizes [43].
and spectral efficiency (SE) are significantly enhanced in
scenarios such as shopping malls, stadiums and subway sta-
tions [25]–[28]. But the ultra dense small cell deployment and B. Caching
coexistence with legacy macro cell raise up the problem of As described above, the spectral efficiency (SE) of wire-
mutual interference reduction, which calls for efficient radio less communication radio access networks has been increased
resource allocation and management technologies [29]–[31]. greatly by ultra-dense small cell deployment. However, this
Meanwhile, Cloud Radio Access Network (C-RAN) is ris- has brought up another issue: the backhaul may become the
ing as a promising solution for reducing inter-cell interference, bottleneck of the wireless communication system under the
CAPital EXpenditure (CapEx) and OPerating EXpenditure circumstance of tremendous and even increasing amount of
(OpEx) [32]. Based on network centralization and virtualiza- data being exchanged within the network [44]. On the other
tion, C-RAN is dedicated to promoting energy efficiency, cov- hand, building enough high-speed backhauls linking the core
erage and mobility performances, while reducing the expense network and the small cells which are growing in number
of network operation and deployment [32]. By pooling the could be exceptionally expensive. Being stimulated by this
baseband resources at BaseBand Unit (BBU), which is located predicament, research efforts have been dedicated to caching
at remote central office (not at the cell sites), C-RAN is able to solutions in wireless communication systems [45]. By storing
provide the network operator with energy efficient operations, Internet contents at infrastructures in radio access networks,
statistical multiplexing gains and resource savings [33], [34]. such as base stations (BSs) and mobile edge computing (MEC)
Massive multiple-input multiple-output (MIMO) technol- servers [46], caching solutions enable the reuse of cached con-
ogy was identified as a key 5G enabler by European 5G tents and the alleviation of backhaul usage. Therefore, the
problem has been transferred by caching solutions from inten- capability. However, due to the size and battery life con-
sive demands on backhaul connection to caching capability of straints, mobile devices tend to fail in fulfilling this demand.
the network. On the other hand, powered by network slicing, SDN and NFV,
However, the original intention of Internet protocols is pro- cloud computing (CC) [54] functionalities begin to incorporate
viding direct connections between clients and servers, while into mobile communication systems, bringing up the con-
the caching paradigm calls for a usage pattern based on dis- cept of mobile cloud computing (MCC) [55], which leverages
tribution and retrieval of content. This contradiction has led to the rich computation resources in cloud for user computation
a degradation on scalability, availability and mobility [47]. To task execution, by enabling computation offloading through
address this issue, Content Delivery Networks (CDNs) [48] wireless cellular networks. After computation offloading, the
and Peer-to-Peer (P2P) networks [49] have been proposed cloud server will create a clone for each user equipment (UE)
in application layer, as first attempts to endow con- individually, then the computation task of the UE will be
tent distribution and retrieval capabilities to the networks performed by the clone on behalf of that UE. Along with
by utilizing the current storage and processing network the migration of computation tasks from UEs to resourceful
infrastructures. cloud, we are witnessing a networking paradigm transforma-
CDNs employ clusters of servers among the Internet infras- tion from connection-oriented networking to content-oriented
tructures and serve the UEs with the replications of content networking, which stresses on data processing, storage and
that have been cached in those servers. The content requests retrieval capability, rather than the previous criterion that
generated by UEs are transmitted through the Domain Name solely focuses on connectivity.
Servers (DNSs) of the CDN to the nearest CDN servers who Nevertheless, due to the fact that the cloud is usually dis-
hold the requested content, in order to minimize the latency. tant from the mobile devices, the low-latency requirements
The decision on which server is chosen to store the replica- of some latency sensitive (real-time) applications may not be
tion of content is made upon a constant monitoring and load fulfilled by cloud computing. Moreover, migration of large
balancing on data traffic in the network [50]. amount of computation tasks over a long distance is some-
P2P networks rely on the storage and forwarding of replica- times infeasible and uneconomical. To tackle this issue, fog
tions of content by UEs. Each UE can act as a caching server computing [56]–[59] has been proposed to provide the UEs
in P2P networks [50], [51]. In P2P networks, the sources of with proximity to resourceful computation servers. The termi-
content are called peers. Instead of caching a complete con- nology fog (From cOre to edGe) computing, was first coined
tent, each peer may only store a part of the content, which is in 2012, by Cisco [60]. It is a distributed computing paradigm
called a chunk. Thus, the content request of a UE is resolved in which network entities with different computation and stor-
and directed to several peers by a directory server. Each peer age abilities and various hierarchical levels are placed within
will provide a part of the requested content upon receipt of short distance from the cellular wireless access network, con-
the request. necting user devices to the cloud or Internet. It is worth noting
Although CDN and P2P do give a solution for content dis- that fog computing is not a replacement but a complement of
tribution and retrieval, due to the fact that they solely operate cloud computing, due to the reason that the gist of fog com-
on application layer and the commercial and technological puting is providing low-latency services to meet the demands
boundaries that they are confined to, the performances of these of real-time applications, such as smart traffic monitoring,
two techniques are not ideal enough to fulfil the demands on live streaming, etc. However, when the applications requir-
network caching services [50]. ing tremendous amount of computation or permanent storage
Serving as an alternative to CDN and P2P networking, are concerned, the fog computing infrastructures are only act-
Information-Centric Networking (ICN) emphasizes on infor- ing as gateways or routers for data redirection to the cloud
mation dissemination and content retrieval by operating a computing framework [57].
common protocol in network layer, which can utilize current The problems in wireless networks such as congestion of
storage and processing network infrastructures to cache con- Internet connection, low-bandwidth and infeasibility of real-
tent. In general, depending on the location of caches, caching time applications can hardly be solved by simply and blindly
approaches of ICN can be categorized as on-path caching boosting the underlying network bandwidth capacity. By ana-
and off-path caching [52], [53]. On-path caching concerns lyzing real data collected from three hotels in Italy, the work
the delivery paths when considering caching strategies, and in [59] shows that fog computing is competent in alleviating
hence are usually aggregated with the forwarding mechanisms the effects of those problems. This analysis demonstrates that
of ICN. On the contrary, off-path caching solely focuses on by deploying fog nodes and corresponding applications at the
the storage and delivery of content, regardless of the delivery network edge, fog computing can proactively cache and man-
path. age up to 28.89 percent of the total traffic, which cannot be
managed by conventional networking and caching approaches.
Moreover, by imposing selective bandwidth limitations on spe-
C. Computing cific connected devices, the fog nodes enable a dynamic and
As the prevalence of smartphones is dramatically growing, flexible management of the available bandwidth, bringing ben-
new mobile applications, such as face recognition, natural lan- efits to the wireless network in terms of resources optimization
guage processing, augmented reality, etc. [1] are emerging. and performance [59]. This real-data-based analysis can serve
This leads to a constantly increasing demand on computational as a preliminary validation of the feasibility and significance
of the incorporation of fog computing into wireless

networks.
Also intended to solve the long-latency problem of mobile
cloud computing, cloudlet based mobile cloud computing is
proposed [61]. Instead of utilizing the computation resource
in a remote cloud, cloudlet based mobile cloud computing
can reduce the data transmission delay by deploying com-
puting servers/clusters within one-hop WiFi wireless access.
However, there are two demerits associated with cloudlet based
mobile cloud computing: First, cloudlet based mobile cloud
computing can hardly guarantee a pervasive service due to
the limited coverage of WiFi access network, which is usu-
ally employed in indoor environment. Secondly, due to the
constraint on infrastructure size, the servers of cloudlet based
mobile cloud computing typically provide small or medium
amount of computation resource, which can hardly meet the
computation requirement in case of large number of UEs.
Fig. 2. Schematic graph of integrated networking, caching and computing.
To serve as a complement of cloudlet based mobile cloud
computing, a new MCC paradigm similar to fog computing,
named mobile edge computing (MEC) is proposed [46], [62].
typical measurement of caching capability is the size of
Being placed at the edge of radio access networks, the MEC
stored information, whose unit is byte. Without alter-
servers can provide sufficient computation resource in physi-
ing the information flow in network, caching alleviates
cal proximity to UEs, which guarantees the fulfilment of the
the backhaul and enhances the capability of information
demand of fast interactive response by low-latency connec-
delivery over a long term.
tions. Therefore, mobile edge computing is envisioned as a
• Computing: The computing vector under consideration in
promising technique to offer agile and ubiquitous computation
this article pertains to its capability to perform algebraic
augmenting services for mobile devices, by endowing con-
or logical operations over the information flows within
siderable computational capability to mobile communication
the network. In contrary with networking and caching
networks [63].
vectors, the operation of computing vector is intended
to alter the information flow in the network. One of the
typical measurements of computing capability is number
III. M OTIVATIONS AND R EQUIREMENTS of information flows or nodes involved in the comput-
OF I NTEGRATION OF N ETWORKING , ing operation, which is named as degree of computing
C ACHING AND C OMPUTING (DoC) [3].
In this section, we first define the integrated system of Given the definitions of networking, caching and computing
networking, caching and computing in question, then we in wireless or mobile systems, we next move on to the discus-
discuss the motivations of the integration, followed by the sion on the topic of this article, integrated networking, caching
requirements of the integration of networking, caching and and computing. In this article, we consider integrated frame-
computing. works and technologies in wireless or mobile communication
systems that take into consideration at least two aspects of
networking, caching and computing as defined above, aiming
A. What Is Integration of Networking, Caching and at providing coherent and sustainable services to fulfil various
Computing mobile network user demands, such as high-data-rate com-
munication, low-latency information retrieval and extensive
In order to give a clear description of integrated networking,
computation resources. In other words, instead of solely focus-
caching and computing, we first give the definitions of
ing on networking vector, we consider solutions concerning
networking, caching and computing considered in this paper,
three network vectors, namely networking, caching and com-
respectively.
puting, as shown in Fig. 2. The acronyms “IoT”, “AR” and
• Networking: The networking vector concerned in this
“SMS” in Fig. 2 stand for “Internet of Things”, “Augmented
article pertains to its capability to deliver data through
Reality” and “Short Message Service”, respectively.
networks with certain bandwidth and power. The mea-
surement of this capability is usually data rate, whose unit
is bit per second. The relationship among data rate, band- B. Why Do We Need Integration of Networking,
width and signal to noise power ratio is well demonstrated Caching and Computing
by Shannon’s capacity formula. 1) The Growth of Networking Alone Is Not Sustainable:
• Caching: The caching vector considered in wireless com- Since the first application of commercial mobile cellular ser-
munication systems pertains to its ability to store a certain vices around 1983, excessive research endeavours have been
amount of data at the infrastructures in the network. The dedicated to the improvement of communication capacity of
mobile communication systems [3]. It is unanimously granted services that rely on intensive computation resources and enor-
that in the early days of mobile communication systems, mous storage capabilities, and thus the growth of networking
when the main functionality of the system is voice and tex- alone is unsustainable. The integration of networking, caching
ting services, the improvement on throughput, latency and and computing is indispensable to achieve sustainability of
link capacity can significantly promote the user experience, mobile communication systems.
which in turn yields greater profits for the network opera- 2) The Benefits Brought by the Integration of Networking,
tors. However, as years passed by, the mobile communication Caching and Computing: The incorporation of caching func-
systems have undergone an evolution, in which the function- tionality into mobile communication systems enables informa-
alities of them have extended greatly. Especially in recent tion distribution and high-speed retrieval within the network.
years, along with the growing prevalence of smartphones, new In-network caching and multicast mechanisms facilitate the
applications are emerging constantly. transformation of data delivery pattern from direct server-client
Due to the fact that many multimedia applications rely on link to information-centric paradigm, thus enhancing the effi-
tremendous amount of data, the demands of mobile network ciency and timeliness of information delivery to users [52].
users on powerful storage services are rapidly increasing these Furthermore, network traffic reduction can also be realized
days. According to the report of ARCchart [64], with the by employing in-network caching in mobile communication
radically increasing camera resolution of mobile devices and systems. According to the trace-driven analysis in [66], with
even faster growing size of data generated by such hardware in-network caching mechanism, three or more hops for 30% of
advances, the demands on mobile cloud storage are undergoing the data packets in the network can be saved, or equivalently
an exponential growth. To meet this user requirement, mobile 2.02 hops on average being saved. Moreover, up to 11% of the
cloud storage has risen and taken the place of most widely user requests for data can be retained within the requesters’
used cloud mobile media (CMM) service, which is mainly pro- domain.
vided by Google, Apple, Dropbox and Amazon today [1]. This On the other hand, the incorporation of computing function-
service enables mobile network users to store video, music and ality into mobile communication systems brings the benefits
other files in the cloud, to access files through any devices irre- of access to the rich computation and storage resources in the
spectively with the source of the data, and to synchronize data cloud or the network edge, which enables the utilization of
in multiple devices. In order to employ mobile cloud storage cutting edge multimedia techniques that are powered by much
service in large scale, it is imperative to ensure high integrity more intensive storage and computation capabilities than what
and availability of data, as well as user privacy and content the mobile devices can provide, thus offering richer and finer
security. It is clear that large scale deployment of mobile cloud multimedia services than what primitive mobile applications
storage service will cause a significant increase in mobile data can provide. Moreover, due to the fact that mobile applica-
traffic, which in turn poses a demand on the integration of tions enabled by MCC/MEC technologies can exploit not only
the storage functionality with ubiquitous and high-data-rate the resources in cloud or network edge, but also the unique
wireless networking functionality. resources in mobile devices such as location information and
On the other hand, audio and video streaming based service, camera data, these applications are apparently more efficient
which typically needs excessive computation resources, is and richer in experience than PC applications.
becoming one of the major services in mobile communication The integration framework makes it possible for users to
systems [1]. According to the technical report of CISCO [65], access media and information through any platform, any
by the end of 2017, the video traffic is expected to take over device and any network. According to Juniper Research,
almost 80% of the annual Internet traffic, whose total size the revenues of consumer cloud mobility services, whose
would exceed 1.4 zettabytes.1 However, the very limited com- initial prototype are cloud based video and music caching
putation resources provided by mobile devices can hardly meet and downloading services like Amazons Cloud Drive and
the requirement raised by tasks such as encoding, transcod- Apples iCloud, have reached 6.5 billion dollars per year
ing and transrating in this category of service. Therefore, to by 2016 [67]. Table I lists some of the applications
solve this dilemma, the incorporation of powerful computing enabled by the integration of networking, caching and
technologies like MCC and MEC into mobile communica- computing [1].
tion systems becomes inevitable. By leveraging services of
MCC and MEC, not only the time consumption of computing
can be reduced, but also the computing and backhaul costs C. The Requirements of Integration of Networking,
can be reduced by caching popular videos through caching Caching and Computing
functionality of MCC and MEC. Furthermore, the elasticity In order to implement the integration of networking, caching
of computation resources provided by MCC/MEC servers is and computing, several requirements need to be met.
suitable for handling volatile peak demands in a cost-effective 1) Coexistence: In an integrated framework, it is impor-
pattern. tant for the caching and computing infrastructures to coexist
In summary, the sole emphasis on networking capabilities with the mobile communication infrastructures, to guaran-
such as connectivity and spectral efficiency can hardly ful- tee the seamless cooperation of the networking, caching and
fill the user demands on the diversity of services, especially computing functionalities. Actually, the coexistence is the pre-
requisite of the realization of integrated networking, caching
1 1 zettabyte=1021 bytes. and computing systems [10].
TABLE I
A PPLICATIONS E NABLED BY THE I NTEGRATION OF N ETWORKING , C ACHING AND C OMPUTING
The infrastructure type, topology, QoS requirement, service existing networking resources. Therefore, from the economic
type and security level are all different across these three point of view, the backward compatibility with legacy com-
functional vectors, and when they are integrated into one munication systems is greatly beneficial and vital. This means
framework, they still need to bear all these different properties. that the compatibilities on both interfaces and protocols should
2) Flexibility: A certain degree of freedom in these three be taken into consideration while designing the integrated
functional vectors is necessary. The flexibility depends on the systems.
degree of the integration of the three vectors [3]. High level of
integration may provide good multiplexing of the three vectors, IV. F RAMEWORKS OF I NTEGRATED N ETWORKING ,
seamlessness of cooperation between different functionalities C ACHING AND C OMPUTING S YSTEMS
and simplicity of implementation, while reducing the flexibil-
ity of resource customization in the integrated system. A low In this section, we summarize the typical frameworks (archi-
level of integration can lead to the reverse effects. tectures) of integrated systems of networking, caching and
computing. The presented frameworks are based on existing
3) Manageability and Programmability: The configuration,
studies, and reflect the basic ideas, mechanisms and com-
deployment and various resources allocation of the integrated
ponents of the integrated systems. Unfortunately, this brief
framework bring up the requirements of manageability and
summary may not cover all of the proposed architectures
programmability [3]. Only with a relatively high degree of
in literature, due to the fact that each proposed architecture
manageability and programmability, can the admission, main-
has its unique original intention, idea of design and working
tenance and resource scheduling of the integrated system be
environment.
viable. To realize programmability, the system needs to pro-
vide effective programming language, feasible interfaces and
secure programming environment with a considerable level of A. Caching-Networking Framework
flexibility. In this subsection, we will stress on the study of [68], due
4) Heterogeneity: Different functional vectors, networking, to the fact that it provides a rough summary of various typi-
caching and computing typically operate in different network cal caching approaches within one framework. Without losing
paradigms, and heterogeneous network structure can facilitate generality, we will also give a quick glance on frameworks
the coexistence of different networking access manners [10]. proposed by other studies.
5) Scalability: Since the number of mobile users and the The framework proposed in [68] intends to provide a dis-
complexity of mobile applications are increasing as time goes tributed caching and content delivery solution, taking into
by, the integrated system of networking, caching and comput- consideration diverse storage, content popularity distribution
ing should have the capability to support a fluctuating number and user device mobility. In this framework, the communica-
of UEs and complexity of tasks. tion and caching capabilities of small cell base stations (SBSs)
6) Stability and Convergence: The integration of and device-to-device (D2D) enabled mechanism are jointly
networking, caching and computing should be stable utilized to support a multilayer content caching and delivery
enough to overcome errors, misconfiguration and other insta- framework. Both SBSs and user devices have the ability to
ble situations. Also, any algorithm concerning the allocation store and transmit content.
of multiple resources should have good convergence property. The multilayer caching-networking framework is illustrated
7) Mobility: The integrated system should be able to sup- in Fig. 3. Multiple small cells are deployed in one macro cell,
port not only the geographical mobility of mobile users, but and all the SBSs are connected to the macro cell base sta-
also the logical mobility of caching and computing functional tion (MBS) through backhaul links. The MBS is connected
vectors within the system. to a gateway of the core network via high-speed interface.
8) Backward Compatibility: Integrated systems being what Each individual SBS and UE is equipped with a caching
they are, should be able to exploit the existing communication device to perform caching function. According to current stan-
network infrastructures up to the hilt. The caching and comput- dards of the Third Generation Partnership Project (3GPP), X2
ing functionalities should be realized on the basis of utilizing interfaces are solely intended for signaling between SBSs, and
Fig. 4. Signaling interaction of the networking-caching framework [68].

Fig. 3. Networking-caching framework [68].
therefore cached content sharing between SBSs is not per- case, the associated SBS will require the demanded content
mitted. However, by utilizing inter-cell D2D communications, from neighbouring SBSs through virtual links, which means
content cached in different small cells can be shared with each these two SBSs are able to communicate with each other via
other. their respective backhauls to the MBS.
As shown in Fig. 3, this framework consists of the following Three function modules, namely REQProxy, ContentServer,
5 cached content delivery approaches. and GateServer, are designed in application layer to enable
1) D2D Delivery (Fig. 3a): Content requested by UEs can the above five content delivery approaches. REQProxy recog-
be retrieved through D2D links from neighbouring UEs. High nizes content requests from UEs and transmits the requests to
density of UEs can contribute to high probability of successful ContentServer. ContentServer performs content retrieving and
retrieval of requested content. Moreover, the implementation forwarding. GateServers are embedded in SBSs, and they send
of inter small cell D2D communications can further facilitate content requests that cannot be handled locally to other SBSs
the utilization of content cached in adjacent small cells. or remote servers.
2) Multihop Delivery via D2D Relay (Fig. 3b): D2D com- The signaling interaction procedures of this framework are
munication links can serve as a relay for content delivery from shown in Fig. 4. In the following, three delivery approaches
other sources to the destination UE. This approach enables a are taken as examples to demonstrate the signaling interaction
broader range of caching and delivery. procedures.
3) Cooperative D2D Delivery (Fig. 3c): If multiple UEs Cooperative D2D delivery (Fig. 3c): When any content is
held the content requested by another UE, they can coop- needed by a certain UE, the Content REQ (content request)
eratively transmit the content to destination UE through a is sent by the UE to nearby UEs, then each individual
D2D multiple-input multiple-output (MIMO) technology to receiver’s REQProxy will contact its own ContentServer to
accelerate the transmission process. Moreover, if video coding check whether the requested content is cached by itself. If
techniques such as multiple description coding and scalable so, the REQProxy of the receiver will respond the REQProxy
video coding were taken into consideration during video data of the requester with a RESP ACK (acknowledgement). After
deliveries, more benefits may be gained. that, the content transmission link will be established with
4) Direct SBS Delivery (Fig. 3d): If the requested content another two packets, Establish Connection and Connection
were cached in the associated SBS, the SBS can transmit ACK. This signaling interaction procedure is also applica-
it directly to the destination UE. Despite the interference ble in the process of plain D2D delivery (Fig. 3a) and direct
incurred during the retrieval and transmission processes, the SBS delivery (Fig. 3d), where the associated SBS responds
latency can be lower than multihop transmission approaches. the requester with RESP ACK.
5) Cooperative SBSs Delivery (Fig. 3e): If the content Cooperative SBSs delivery (Fig. 3e): In this approach, the
requested by a certain UE were stored neither in its nearby requester UE sends a Content REQ to its associated SBS,
UEs nor in its associated SBS, it may retrieve and demand then the SBS will retransmit this Content REQ to the MBS,
the content from a SBS in a neighbouring small cell. In this through GateServer. The virtual link between the SBS holding
service providers [72]. In this approach, several services share

a number of physical contents cached in the network, and
each content is sliced into a number of virtual contents. Each
service can utilize one or several of the virtual contents with-
out the knowledge of other virtual contents. The number of
virtual copies of the physical contents can depend on the pop-
ularity of the content, i.e., more popular physical content is
sliced into more virtual contents. Conceptually speaking, the
content-level slicing is an realization of caching sharing and
dynamic access in virtualized networking environment.
Not only the reduction of network data traffic but also
the improvement of quality of experience can be realized
Fig. 5. The information-centric networking [8]. by moving the frequently requested content toward network
edge closer to UEs [71]. The study of [73] proposes a frame-
work of caching popular video clips in the RAN. The concept
the requested content and the requester’s associated SBS will of FemtoCaching is proposed in [74], and the video content
be established. Then the content will be transmitted to and caching and delivery approach with the facilitation of dis-
cached at the associated SBS and eventually delivered to the tributed caching helpers in femtocell networks is discussed.
requester UE. The research conducted in [71] studies current caching tech-
Multihop delivery via D2D relay (Fig. 3b): In this approach, nologies and discusses caching in 5G networking, including
the relay UE needs to establish a link with its associated SBS. radio access network caching and evolved packet core network
When the requested content is ready to be transmitted, the caching. Then an edge caching approach (on edge of RAN)
SBS will send a connection ACK to the relay UE in order to based on information-centric networking is designed. The
establish transmission connection. paper discusses about architecture design issues on caching
An important feature of 5G mobile networking could within evolved packet core (EPC), caching at RAN and
be recognizing the content as the first-class citizen in the content-centric networking (CCN) based caching.
network [8]. To fulfil this requirement, the conventional con- As the popularity of social media is increasing dramatically,
tent delivery networking (CDN) [69], [70] has distributed online social networks (OSN) are starting a new revolution on
further, incorporating content aware and information-centric the manners that people communicate and share information.
caching technologies, and forming the so called information- In centralized OSN, the network has a central repository of
centric networking (ICN) [71], who has the significant advan- all user data and imposes control regarding how user data
tages of natively facilitating efficient and scalable content will be accessed. This form of architecture has the advan-
retrieval and enhancing mobility and security of the network. tage of efficient dissemination of social updates. However,
As an important native feature of ICN, in-network caching the users can hardly gain control over their personal data,
has the capability of significantly reducing duplicate data and are therefore constantly exposed to potential privacy vio-
transmissions in the network. The ICN based air caching tech- lations. To address this issue, the architecture of distributed
nology is considered as a promising candidate to support the online social networks (DOSN) has been proposed [75], [76].
implementation of the SDN based 5G mobile network [71]. DOSN does not have a central repository and hence guar-
The study of [8] proposes an information-centric wire- antees the user privacy, but faces the problem of inefficient
less network (ICN) virtualization framework, which involves dissemination of social updates. The work in [75] introduces
content-level slicing functionality. As shown in Fig. 5, in con- Social Caches as a way to alleviate the peer-to-peer traffic
trast with Internet Protocol (IP), which adopts a host-based in DOSNs, and the social caches are supposed to perform as
conversation model (i.e., establish and maintain connection local bridges among friends to provide efficient information
between hosts before actual data delivery) and follows a source delivery. Furthermore, a social communication model called
driven data delivery approach (i.e., path is initiated by sender Social Butterfly is proposed to utilize social caches. However,
to receiver), ICN follows a receiver driven communication the selection of social caches demands full knowledge of
principle (i.e., path is initiated by receiver to sender), and the the whole social graph. Therefore Han et al. [76] propose a
data transmission follows the reverse way. As can be seen in number of distributed social cache selection algorithms.
Fig. 5, network nodes in ICN can cache the content that flows Due to the fact that stable D2D links can be easily
through it and disseminate the content to other UEs. Instead of maintained among people who have close social relation-
the reachability of the hosts and the maintenance of the con- ships, social-aware device-to-device (D2D) communication
nectivity between hosts, the main concern of ICN is retrieval is proposed to complement the online social network com-
and dissemination of information. munication. The work in [77] proposes a fog radio access
Liang et al. [8] present a content-level slicing in their network (FRAN) supported D2D communication architecture,
proposed virtualized ICN framework, which can be considered in which the UEs can download files either from centralized
as an extension of dynamic content access and sharing. The baseband unit (BBU) pool through distributed remote radio
physical content (cache) is sliced through space multiplexing, heads (RRHs), or from other UEs through D2D links. In [78],
time multiplexing and so forth, then allocated to different a context-aware framework for optimizing resource allocation
Fig. 6. Cloud mobile media framework, demonstrating data and control flows [1].
in small cell networks with D2D communication is proposed.

Both the wireless and social contexts of wireless users are
utilized to optimize the resource allocation and improve the
traffic offload.
B. Computing-Networking Framework
In this subsection, we will consider the frameworks of Cloud
Mobile Media (mobile cloud computing) and mobile edge
computing as two representatives to give a general profile of
the architecture of computing-networking paradigm.
1) Cloud Mobile Media: The overall architecture of cloud
mobile media (CMM) is depicted in Fig. 6, including end to
end control and data flows between Internet cloud servers and
mobile devices [1]. Enabled by cloud mobile media, appli-
cations on mobile devices are able to utilize intensive and
elastic caching and computing resources in remote clouds (typ-
ically located in the Internet), including private, public and Fig. 7. MEC server platform overview [46].
federated (hybrid) cloud types. The user interfaces of CMM
applications used for commands input are provided by user
They both allow UEs to offload computation tasks to servers
devices, including touchscreens, gestures, voices and texts.
that execute the tasks on behalf of the UEs, then the results are
The subsequent control commands are then transmitted uplink
returned to the UEs. The architecture of MEC server platform,
via cellular radio access networks (RAN) to gateways located
which consists of a hosting infrastructure and an application
in core networks (CN), and eventually to the serving clouds.
platform, is summarized in Fig. 7 [46].
After that, the clouds will perform data processing or con-
The MEC servers can be deployed at various locations
tent retrieval using computing or caching resources in cloud
within the edge of radio access networks. According to the
servers, in accordance with the user control commands. Then
report of [46], the MEC server can be deployed at the macro
the results are sent back to user devices downlink through CN
base station (eNB) site, at the multi-technology cells aggre-
and RAN.
gation site, or at the radio network controller (RNC) site.
Please note that most of the applications follow the data and
The deployment scenarios are depicted in Fig. 8. The multi-
control flows shown in Fig. 6, but there exist some exceptions
technology cells aggregation site can be located either in
that slightly deviate from them. For instance, the control flows
indoor environment in an enterprise, such as university center
in cloud based media analytics are not always initiated by
or airport, or in outdoor environment to control several local
mobiles devices, and this type of application may collect data
multi-technology access points providing coverage over cer-
from both the cloud and the mobile devices in order to perform
tain public premises. Direct delivery of fast, locally relevant
analytics for other types of CMM applications.
services from base station clusters to UEs can be enabled by
2) Mobile Edge Computing: As described in previous sec-
this option of deployment.
tions of this article, MEC serves as a complement for mobile
cloud computing, and the intention of MEC is reducing the
high data transmission latency of MCC by providing proximity C. Caching-Computing Framework
to computation resources at the edge of radio access network. In recent years big data has attracted more and more atten-
The basic mechanism of MEC is quite similar to that of MCC. tion due to increasing diversity of mobile applications and
Cleansing, parsing and formatting data: In order to per-

form statistical analysis and machine learning, the data should
be cleaned in advance. The malfunctioning, inappropriate and
inconsistent packets involved in the raw data should be elim-
inated before any type of data processing. Then the relevant
fields should be extracted from the raw data. Finally, the parsed
data should be encoded accordingly.
Data analysis: A variety of data analysis technologies
can be applied on analysis of the header or/and payload
information of both control and data planes. The purpose
of data analysis is finding the relationship between con-
trol/data packets and the requested content and then predicting
spatial-temporal user behaviour.
Statistical Analysis and Visualizations: The outcomes of
the data analysis can be stored and reused for further sta-
tistical analysis. Moreover, the outcomes can be formatted as
graphs or tables for enhancement of readability and ease of
comprehension.
Tang and Quek [80] propose a cloud content-centric mobile
networking (CCMN) architecture, which incorporates cloud
computing into both core network caching and radio access
network caching. In this architecture, the conventional content
delivery network (CDN) migrates into the cloud, forming a
cloud CDN which leverages the merits of both cloud com-
Fig. 8. MEC server deployment scenarios [46]. puting and CDN. As an application of cloud computing, the
cloud CDN architecture inherits most of the advantages of
cloud computing, and moreover, it can offer sufficient elas-
tremendous amount of data stream in mobile networks. In this ticity for handling flash crowd traffic. However, the content
context, data management/processing and data caching have proximity to UEs and the coverage of cloud CDN is not as
emerged as two key issues in this field. The study of [79] ideal as those of conventional CDN.
proposes a caching-computing framework in which cached The mergence of cloud computing and RAN forms the so
content is moved from cloud to RAN edge, and a big data called cloud RAN (C-RAN), whose most significant innova-
platform powered by machine learning and big data analytics tion is that the computationally intensive baseband processing
is deployed in core site for content selection. tasks are executed by a centralized cloud baseband unit (BBU)
As illustrated in Fig. 9, the caching platform at small cell pool, instead of the traditional distributed baseband processing
base stations (SBSs) and the computation and execution of devices at the base station sites [81]. By adding the con-
the content prediction algorithms at the core site are two key tent caching functionality into conventional C-RAN, C-RAN
components of the proposed architecture. A big data platform with caching as a service (CaaS) is considered as a promising
is employed at the core site to perform tracking and prediction approach to overcome the limitations of core network caching,
of users’ demands and behavior for content selection decision, due to the fact that the content is moved from core network
and SBSs in RAN are equipped with caching units to store the to RAN, and hence is much closer to UEs. In C-RAN with
strategic content selected by the big data platform. The reason CaaS, the BBU pool can offer larger caching space than that
of employing the big data platform is that the content cached in of BSs, and the centralized BBU pool can facilitate solving
SBSs can hardly cover all the users’ content demands, due to the distributed content placement problem over geographically
limited storage capacity of the SBSs. In order to yield optimal scattered BSs. Nevertheless, C-RAN with CaaS doesn’t do
performance, it is important to select the most popular content well in fronthaul traffic mitigation.
(most frequently requested content based on estimation) and The overall architecture of the cloud content-centric mobile
to decide the content cache placement at specific small cells networking (CCMN) is depicted in Fig. 10, which shows the
based on backhauls, rate requirements, content sizes and so assistant role of cloud computing in both core network caching
forth. and RAN caching.
The big data platform portrayed in Fig. 9 is used for sort- A similar framework called fog RAN (F-RAN), com-
ing users’ data traffic and extract statistical information for bining edge caching and C-RAN, has been recently advo-
proactive caching decisions. To achieve this goal, the following cated [82]–[84]. In F-RAN, edge nodes are equipped with
requirements should be met by the big data platform. caching capabilities, and are simultaneously controllable to a
Enormous amount of data processing in short time period: central cloud processor like in C-RANs. C-RANs can guar-
Making proactive caching decisions requires the capability antee high spectral efficiency due to cooperative cloud-based
of processing large amount of data and drawing intelligent transmission, while the latency is not ideal because of fron-
insights in short time. thaul transmission. On the other hand, edge caching can
Fig. 9. Caching-computing architecture [79].
in SDN, which facilitates to guarantee flexibility of the

solution. Three types of radio access networks are consid-
ered in this paper, namely cellular networks, wireless local
area networks (WLANs) and worldwide interoperability for
microwave access (WiMAX). Each network node is equipped
with caching and computing capabilities, and is under the con-
trol of a controller, which is in charge of the topology of the
Fig. 10. Architecture of cloud content-centric mobile networking [80]. whole wireless network and the packet forwarding strategies.
By doing so, this framework can provide the UEs with caching,
computing, as well as communication services.
provide low-latency deliveries of popular content, but suffers 2) Networking and Computing Assisted Caching: Caching
from poor interference mitigation performance due to decen- systems are facing three primary issues: where to cache, what
tralized baseband processing at the edge nodes. By leveraging and how to cache, and how to retrieve. To answer the ques-
features of both edge caching and C-RAN, F-RAN can bene- tion where to cache, the identification of the eligible candidates
fit from both low-latency deliveries of content and centralized for caching the content is needed. In other words, it is nec-
baseband processing. essary to clarify the criteria of a node being an information
hub in the network. Following that, the question what to cache
with respect to the popularity and availability of the content
D. Caching-Computing-Networking Framework is posed. It is also imperative to decide which nodes should
In this subsection, we generally consider two cate- keep which content for redundant caching avoidance. After
gories of networking-caching-computing frameworks, namely the content is appropriately cached in the network, the final
networking-caching-computing convergence and networking question how to retrieve requires an answer.
and computing assisted caching. The former one refers to the In order to accommodate the dramatically increasing content
framework in which networking, caching and computing func- requests generated by huge number of UEs in geographical
tionalities coalesce with each other to form a comprehensive proximity in vehicular networks, the study in [85] advocates
structure, providing the UEs with versatile services. The later the adoption of a networking and computing assisted caching
one means in this type of frameworks networking and com- paradigm in information-centric networking architecture, for
puting capabilities of the network elements are functioning ICN can provide the decoupling of content providers and con-
to provide supports to caching functionality, improving the sumers, and in-network caching functionality is intrinsically
quality of caching services of the networks. supported by ICN. Therefore, the paper proposes a socially
1) Networking-Caching-Computing Convergence: The syn- aware vehicular information-centric networking model, in
ergy of networking, caching and computing is not a trivial step. which mobile nodes such as vehicles are equipped with com-
First, different resource accesses require unique communica- munication, caching and computing capabilities, and they are
tion protocols, and therefore the convergence of these three responsible for computing the eligibility of themselves for
types of resource accesses poses a requirement on new pro- caching and delivering content to other nodes (i.e., the rela-
tocol design. Secondly, interactions between different types tive importance of vehicles in the network, with respect to user
of resources need appropriate interfaces on network devices, interests, spatial-temporal availability and neighborhood con-
which means redesign of network devices is necessary. nectivity). Furthermore, each node also bears the mission of
Huo et al. [5] propose a framework incorporating computing the eligibility of itself for being an information hub
networking, caching and computing functionalities under the in the network. This has addressed the where to cache ques-
control and management of a SDN control plane, aiming at tion. For the what and how to cache question, the decision is
providing energy-efficient information retrieval and comput- made on the computation results regarding content popularity,
ing services in green wireless networks. A software-defined availability and timeliness, taking into consideration cooper-
approach is employed in the integrated framework, in order ative caching schemes. As for the how to retrieve question,
to leverage the separation of control plane and data plane the paper presents a social content distribution protocol for
Fig. 11. A use case of the integrated system of networking, caching and computing.
vehicles to relay and retrieve cached content, in order to

enhance the information reachability.
E. A Use Case
Here we provide a use case to further elaborate the interac-
tions of the three dimensions in the integrated framework.
In this use case, we consider wireless network virtualiza-
tion (WNV) technology, which realizes physical resources
sharing between different virtual networks by creating virtual
BSs upon physical infrastructures. A virtual BS could be a
BS, access point (AP) or road side unit (RSU) in the physical
wireless network. As shown in Fig. 11, a cluster of virtual BSs
are connected to one MEC server, and they are all connected
to the Internet through the core network. A cache (storage
facility) is deployed at each virtual BS.
Joe finds an interesting monument when he is visiting
Ottawa. With his augmented reality (AR) glasses, Joe notices
that there is an introductory video about this monument. Being
interested in this video, he generates a video content request Fig. 12. The processing procedure of the use case.
via his AR glasses to a virtual BS. According to the descrip-
tion of the video content and the information about the user,
the virtual BS will check whether its associated cache has the he takes a video of it, puts some comments on it, and streams
requested content. If yes, the virtual BS will further exam- it to a remote server via a virtual BS. The virtual BS extracts
ine if the version of its cached video content matches the the video content, and estimates the possibility that other peo-
user’s device. If still yes, the virtual BS will directly send ple want to see this video. If the possibility is high, the virtual
the requested content to the user from the cache. If no, the BS will cache this video locally, so that other people who are
virtual BS will extract the current video content and initi- interested in this video can obtain this video from its cache
ate a computation task according to the size of the content directly, instead of requesting it from the remote server. As
including the input data, codes and parameters, as well as the such, the backhaul resources can be saved and the propagation
number of CPU cycles that is needed to accomplish the com- latency can be reduced. If the cached version does not match
puting/transcoding task. Then the virtual BS will transmit the with the new user’s device, transcoding will be performed by
task to the MEC server to execute the computation. After the the MEC server.
computation is finished, the virtual BS sends the transcoded Although the above use case focuses on tourism services,
video content to the user. If the virtual BS cannot find any ver- similar use cases with similar requirements can be derived
sion of the requested video content in the cache, the virtual for other services and applications, such as transportation,
BS has to retrieve the content from the Internet, and this will education, healthcare, etc.
inevitably consume some of the backhaul resources. Upon the
arrival of the new video content at the virtual BS, the virtual
BS can choose whether to store the content in its cache or not. V. P ERFORMANCE M ETRICS OF I NTEGRATED
The procedure is shown in Fig. 12. N ETWORKING , C ACHING AND
The above procedure is for downlink streaming video. C OMPUTING S YSTEMS
Similarly, the procedure for uplink streaming video can be con- In this section, we present the performance metrics of inte-
sidered as follows. When Joe finds an interesting monument, grated networking, caching and computing systems. We first
TABLE II
P ERFORMANCE M ETRICS OF N ETWORKING -C ACHING -C OMPUTING I NTEGRATED S YSTEMS
present the general metrics of the entire integrated system, 3) Recovery Time: The recovery time metric is an indicator
then we present the networking, caching and computing related of the resilience of the networking-caching-computing inte-
metrics respectively. All the metrics are classified and briefly grated system, which refers to the time duration in which the
described in Table II. system recovers from a certain type of failure or attack [87].
A. General Metrics B. Networking-Related Metrics

1) Cost: The cost metric refers to the investment paid by 1) Coverage and Capacity (Throughput): The coverage
infrastructure providers and system operators for infrastruc- metric refers to the entire geographical area that is covered
ture construction and system operation, including CapEx and by the radio access network service, through which users can
OpEx. For networking-caching-computing integrated systems, gain access to communication, caching and computing services
the CapEX is composed of the expenses of constructing provided by the integrated system. While capacity refers to the
base stations, radio network controller equipments, backhaul maximal aggregated peak rate (maximum theoretical through-
transmission equipments, core network equipments, caching put) of a certain territory served by a base station or access
and computing equipments installation in network nodes. point [14]. Throughput refers to the aggregated data rate of
While OpEx consists of energy charge, equipments lease and the radio access network in a certain period of time and area.
operation and maintenance fees [86]. The deployment cost Coverage and capacity is one of the fundamental and crucial
includes the lease paid for renting licensed spectrum issued metrics in radio access network design and optimization, and
by authorities. is typically determined by bandwidth, transmitting power and
2) Revenue: Profit, which is defined as the difference network planning, which are succinctly described as follows.
between revenue and cost, can be used to reflect the effec- a) Bandwidth: Bandwidth is the whole radio spectrum in
tiveness of networking-caching-computing integrated systems. a certain area that is available to mobile devices for accessing
Another indicator, revenue to cost ratio (RCR) can also realize the communication, caching and computing services of the
the similar function. integrated system.
b) Transmitting power: Transmitting power is the power • Queuing Delay: Queuing delay refers to the amount of
with which base stations and mobile devices transmit signals time a data packet spends sitting in a queue and awaiting
to each other. transmission due to over-utilization of the outgoing link.
c) Network planning: Network planning refers to topo-
logical design, synthesis and realization of the network, aiming
at ensuring that the new designed network or service can fully C. Caching-Related Metrics
accommodate the demands of the network operators and end
users. 1) Average Latency: In caching-networking systems, the
2) Deployment Efficiency: The deployment efficiency (DE) average content-delivery latency experienced by end users is
metric is defined as the ratio of system throughput or cov- a major performance metric. Typically, there are three types of
erage to deployment costs, which include both CapEx and average latency considered in [91]. The first type is the average
OpEx [86]. Typically, the DE metric is taken as an impor- latency of delivering content to end users from clients’ clos-
tant network performance reference when conducting network est network nodes (such as routers) that cache the requested
planning. content locally. The second type is the average latency of deliv-
3) Spectral Efficiency: The spectral efficiency (SE) metric ering content to end users from peer network nodes. In this
is defined as the fraction of system throughput or cov- case, the requested content is not cached in the network node
erage to bandwidth. SE is usually taken as an important that is directly connected to the client but can be fetched from
criterion in wireless network optimization [86]. Furthermore, a peer network node in the same administrative domain. Note
more detailed SE metrics can further facilitate evaluating that this average latency includes the latency of content trans-
more specific network performances, for instance, the worst mission from peer node to client associating node, and the
%5 users spectral efficiency and the cell edge spectral latency of transmission from associating node to client. The
efficiency. third type is the average latency of delivering content from the
4) Energy Efficiency: The energy efficiency (EE) metric is content origin. These three types of average latency collec-
defined as the ratio of system throughput or coverage to energy tively reflect the average latency incurred by content delivery
consumption. It is worth noting that the energy consumption in the network. The latency metric is also referred to as delay,
includes not only the signal transmission power consump- download time or Round-Trip Time (RTT).
tion but also any other aspects of energy consumption in the 2) Hop-Count: The hop-count stands for the number of
entire network, including equipments energy and accessories, network nodes that are traversed (inspected) by a content
etc. [88]. request before it is satisfied. In order to reflect the network
5) QoS: The quality of service (QoS) metric is generally traffic reduction brought by caching, the hop-count metric is
represented by several variables that characterize the network always determined as a fraction, given the name hop-count
performances experienced by end users [89]. For example, in ratio, hop-reduction ratio or hop-count rate, and is cal-
3GPP LTE [90], the QoS metric is classified into 9 types of culated as the ratio of number of traversed nodes with
so called QoS class identifiers (QCIs), which are associated caching being conducted to the number of traversed nodes
to priority (9 levels), resource types (guaranteed bit rate or without caching being conducted. This metric can be con-
non-guaranteed bit rate), packet loss rate and packet delay sidered as one of the estimations of the delay metric in the
budget. network [45].
6) Signaling Delay and Service Latency: The signaling 3) Load Fairness: The load fairness metric is defined as a
delay metric refers to the propagation delay of control signals ratio of the number of content requests and content responses
between network entities that perform the network manage- received by one node to the total number of content requests
ment functions. and content responses generated in the whole network. This
The service latency metric refers to the delay induced by metric is intended to indicate the distribution of the traffic and
preparation and propagation of data packets in the system. the existence of overloaded nodes.
Since this article only focuses on the radio access networks 4) Responses Per Request: This metric stands for the
in the networking part, the technologies discussed in this arti- number of content responses being received by the content
cle only involve three types of latencies that are specified as requester for one content request. This metric is intended
follows. to estimate the network traffic introduced by the multi-path
• Propagation Delay: As the primary source of latency, mechanism.
propagation delay is defined as a function of how long it 5) Cache Hits: The cache hits metric indicates the load of a
takes information to travel at the speed of light in wireless node by measuring the number of content requests responded
channels from origin to destination. (satisfied) by the node. When intended to reflect the load sav-
• Serialization Delay: Serialization is the conversion of ings of a node due to conducting caching, the node hits metric
bytes (8 bits) of data stored in a device’s memory is determined as a fraction called cache-hit ratio, which is nor-
into a serial bit stream to be transmitted over the malized by the total number of content requests generated in
wireless channels. Serialization takes a finite amount the whole network. This server-hit ratio is also referred to as
of time and is calculated as: Serialization delay = cache-hit probability or cache-hit rate.
packet size in bits / transmission rate in bits per 6) Caching Efficiency: Caching efficiency is defined as the
second. fraction of the number of cached contents that have been
requested during a certain period of time in a node to the VI. E NABLING T ECHNOLOGIES OF I NTEGRATED
total number of contents being stored in the same node. N ETWORKING , C ACHING AND
7) Caching Frequency: Caching frequency is defined as the C OMPUTING S YSTEMS
fraction of the number of cached contents in a node to the total In this section, a number of the enabling technologies for
number of contents that have flowed through the same node integrated networking, caching and computing systems are
during a certain period of time. summarized. These technologies are presented in taxonomy
8) Cache Diversity: The cache diversity metric refers to the based on the types of integrations.
number of distinctive contents cached in the whole network.
9) Cache Redundancy: The cache redundancy metric refers
to the number of identical contents cached in the whole A. Caching-Networking
network. Cache redundancy and cache diversity are the com- 1) Caching in Heterogeneous Networks: As a key com-
plements to one another. ponent of 5G cellular communication systems, heteroge-
10) Absorption Time: The absorption time metric refers to neous networking has been excessively studied, and caching
the time period during which the cached content remains in approaches and corresponding caching gains are well investi-
the node. This metric is determined by the cache replacement gated. A number of typical caching schemes and supporting
policies and content popularity distribution. It is also referred mathematical tools are introduced and discussed here.
to as caching time. A dynamic caching paradigm in hyper-dense small cell
networks is proposed in [101] to address the caching issue
D. Computing-Related Metrics in realistic scenarios. This paper poses the problem that the
1) Execution Time: The execution time metric refers to scenarios considered in literature are always static environ-
the time consumed by the execution of the computation ment, in which mobile users are assumed to remain stationary
task [92]–[96]. In other words, this metric represents the time during data transmitting processes so that the downloading
period during which the computation is executed by network procedure can always be completed by associated small cell
nodes, either user devices or servers. In mobile cloud com- base stations. In order to address the caching problem more
puting and mobile edge computing systems, the time during realistically, the proposed caching paradigm emphasizes on
which the computation task is transmitted to the MCC or a dynamic environment in which mobile devices connect to
MEC server may also be involved in this execution time met- multiple small cell base stations intermittently, and therefore
ric [63], [97]. The data transmission time depends on network the mobile devices may only download part of the cached data
conditions such as bandwidth and interference level, instead due to the limited connecting durations. The failed transmitting
of computing capabilities of network nodes. The execution tasks are then redirected to macro cell base station. Aiming
time metric is also referred to as execution latency or response at minimizing the load of macro cell, this work uses random
latency [98]. walks on a Markov chain to model mobile user movements in
2) Energy Consumption: The energy consumption metric an optimization framework. The proposed caching paradigm
refers to the energy consumed by network nodes, either user leverages the combination of user mobility predictions and
devices or servers, for the execution of the computation task. information-mixing methods based on the principle of network
The energy consumption metric is in close relation to the exe- coding.
cution time metric, since energy consumption depends on the An algorithm is proposed in [102] to jointly consider
time period during which the network nodes work. In MCC caching, routing, and channel assignment for video delivery in
and MEC systems, the energy consumed by user devices for coordinated small-cell wireless cellular systems of the future
computation task offloading, and the energy consumed by the Internet, aiming at maximizing the throughput of the system.
MCC or MEC servers for computation results feedback, can This work stresses on three major issues. First, collaborations
also be included in this energy consumption metric [63], [97]. between small cells are taken into account in cached content
3) Computation Dropping Cost: In some cases, due to var- delivery. When a content request cannot be satisfied by the
ious reasons, the computation task cannot be executed but user’s associated small cell base station, instead of redirect-
have to be dropped. For instance, the task is too computa- ing the request to macro cell base station immediately, the
tionally intensive to be executed on mobile device, while the request is forwarded to other small cell base station to search
MEC server is already overloaded or the wireless link between for possible satisfaction. Secondly, the wireless transmission
mobile device and the MEC server is undergoing deep fading. interference between cache-to-user communication and other
In this case, users’ quality of experience (QoE) can deteriorate links transmitting is taken into consideration to optimize the
due to computation request not being satisfied. Therefore the throughput. Thirdly, cached content delivery routings are han-
computation dropping cost is a penalty associated with this dled differentially depending on the geographical locations of
failure [99]. the mobile users. This work formulates the above described
4) Throughput: The throughput metric refers to the data issues into a linear program. The column generation method is
units delivered through the network per unit time interval. applied to solve the problem by splitting it into subproblems,
This metric is often used in mobile cloud computing and and an approximation algorithm is utilized to further reduce
mobile edge computing systems to serve as a joint indicator the complexity of the problem.
of the network transmission capability and the network nodes Das and Abouzeid [103] consider cooperative caching and
computation capability [100]. content delivery in a heterogeneous network, in which the
spectrum is shared between one macro cell and a number of bands for proactively caching video contents. The harvested
small cells. The cooperation among cells is realized by the bands allocation issue is formulated as a Markov decision
operation that small cell base stations cache primary files that process with hidden and dynamic parameters, which is fur-
are requested by macro cell users, so that some of the content ther transformed into a partially observable Markov decision
requests of the macro cell users can be satisfied by small cell process and a multi-armed bandit formulation. A spectrum
base stations. Two caching and delivery algorithms are devel- management mechanism is designed accordingly to maximize
oped in this cooperative caching scenario. In the first algorithm the revenue of proactive video caching and the efficiency of
the priority of access of the primary files is the same as that spectrum utilization.
of secondary files (files that are for small cell users), while in Li et al. [91] investigate the problem of tradeoffs
the second algorithm the priority of the primary files is higher between in-network storage provisioning cost and the network
than that of secondary files. Due to cooperation mechanism, performance in content-centric networks. A holistic model
both of the algorithms can maintain a greater set of supportable is formed to characterize the cost of global coordina-
request generation rates than that of non-cooperative solutions. tion of in-network storage capability and the network
It is conspicuous and unanimously agreed that caching performance of delivering content to users. The strat-
in small cell networks is beneficial to relief of backhaul egy for provisioning the storage capability that jointly
and improvement of throughput, but it remains ambiguous optimizes the network cost and performance is derived
whether caching is helpful to the improvement of energy effi- accordingly.
ciency (EE) of the network. Therefore, the work in [104] 3) Caching in D2D Networking: As an important com-
explores the impact of caching on EE in small cell networks. ponent of 5G wireless cellular networks, device-to-device
The closed-form expression of the approximated EE is derived. (D2D) communication has attracted extensive attention from
Based on that, the conditions under which the EE can bene- researchers. A major functionality of D2D communication is
fit from caching are given, and the optimal cache capacity content sharing among mobile devices, and therefore caching
that can maximize the EE is deduced. Furthermore, the max- strategy is essential in D2D technique.
imal EE gain brought by caching is analyzed accordingly. Malak et al. [108] propose a content caching scheme in
The major observation of their exploration is that caching D2D networking to maximize the probability of successful
at small cell base stations can benefit the network EE with content delivery, which is derived by using stochastic geom-
power efficient cache hardware being employed, and that if etry in presence of interference and noise. An optimization
local caching already provided EE gain, caching more content problem is formulated to search for the caching distribution
may not further benefit the network EE. that can maximize the density of successful receptions (DSR).
2) Caching in Information-Centric Networking: In order to The content requests are modeled as a Zipf distribution.
address the dramatic growth of mobile data traffic and the Zhang et al. [109] also study the caching scheme in D2D
shift of mobile users’ service preference from peer-to-peer networking. Unlike the work in [108], power allocation is
data transmission to information dissemination oriented ser- taken into consideration in this work. An optimization problem
vices, information-centric networking has risen to serve as a in which D2D link scheduling and power allocation are jointly
promising solution. considered is formulated, aiming at maximizing the system
A collaborative in-network caching solution with content- throughput. Furthermore, the problem is decomposed into an
space partitioning and hash-routing in information-centric optimal power allocation problem and a D2D link-scheduling
networking is studied in [105]. By partitioning the content problem, in order to tackle the non-convexity of the original
space and assigning partitions to caches, the path stretch problem. The D2D link-scheduling problem is solved by an
incurred by hash-routing can be constrained by the proposed algorithm maximizing the number of D2D links under the con-
scheme. The issue of assigning partitions to caches is formu- straints of the transmitting power and the signal to interference
lated as an optimization problem aiming at maximizing the plus noise ratio, while the power allocation problem is solved
overall hit ratio, and a heuristic algorithm is developed to solve by another algorithm maximizing the minimum transmission
the optimization problem. Furthermore, the partitioning pro- rate of the scheduled D2D links.
portion issue is formulated as a min-max linear optimization Unlike the works in [108] and [109], which solely focus
problem, in order to balance cache workloads. on D2D caching schemes, the work in [110] studies the issue
Yuan et al. [106] focus on the encrypted in-network caching of jointly caching and transmitting content in both small cell
in information-centric networking. In order to address the networking and D2D networking. This work considers two
newly raised user privacy exposure and unauthorized video types of caching gains, namely pre-downloading and local
access issues, this work presents a compact and encrypted caching gains, and these two types of gains have an inher-
video fingerprint index to enable the network to locate ent tradeoff between each other. In this context, a continuous
the cached encrypted chunks for given encrypted requests. time optimization problem is formulated, aiming at the deter-
Furthermore, for video delivery purpose, a secure redundancy mination of optimal transmission and caching policies that
elimination protocol which leverages the cached encrypted minimize a generic cost function, such as throughput, energy
chunks is designed. and bandwidth. It is then demonstrated that caching files at
In [107], a dynamic approach is proposed in information- a constant rate offers the optimal solution, due to the fact
centric cognitive radio networks (IC-CRNs) to improve the that it allows reformulation of the original problem as a
performance of video dissemination, by leveraging harvested finite-dimensional convex program.
4) Others: Some approaches that specify other network The framework aims at delivering desirable sensory data to
architectures are introduced here. the mobile users fast, reliably and securely, prolonging the
Liu and Lau [111] study the caching assisted video WSN lifetime, reducing storage requirements, and decreas-
streaming in multicell multiple-input multiple-output (MIMO) ing traffic load. Moreover, the framework can monitor and
networks. A proportion of the video files are cached in base predict the future trend of the sensory data traffic. In [115],
stations and these cached files opportunistically induce Coop- an integrated WSN–MCC architecture is proposed, in which
MIMO transmission, which can achieve MIMO cooperation the cloud is considered as a virtual sink with a number of sink
gain without expensive payload backhaul. In this context, the points, and each of the sink points gathers the data from sen-
approximated video streaming performance is derived by using sors within a zone. Thus, the sensor data is eventually stored
a mixed fluid-diffusion limit for the playback buffer queueing and processed in a decentralized manner in the cloud. The
system. Based on the performance analysis, the cache control work in [116] considers the collaborative location-based sleep
and playback buffer management issues are formulated as a scheduling schemes in integrated network of MCC and WSN.
stochastic optimization problem. Then a stochastic subgradi- In order to address the mobile users’ current location depended
ent algorithm is employed to obtain the optimal cache control, specific data requests and decrease energy consumption of
and a closed form solution for the playback buffer thresholds the sensors, the proposed scheme dynamically determines the
is given. awake or asleep status of sensors based on the current locations
Since the economic viability of cellular networks is being of the mobile users.
challenged by the dramatically increasing mobile data, the Collaborative task execution between mobile devices and
work in [112] studies the solution of leasing wireless band- mobile cloud servers relies on the quality of wireless commu-
width and cache space of access points (APs) in residential nication provided by wireless networks. The research of [117]
802.11 (WiFi) networks, in order to reduce cellular network investigates this issue under a stochastic wireless channel.
congestion and relieve the backhaul link payload of the APs. The task offloading and collaborative task execution strategies
Monetary incentive (compensation) is provided to APs, in pro- are formulated as a constrained shorted path problem, aiming
portion to the amount of bandwidth and cache space that they at minimizing the energy consumption of mobile devices. A
are leasing, which in turn influence the caching and rout- one-climb policy and an enumeration algorithm are proposed
ing policies. In this context, a framework taking into account to solve the problem. Furthermore, the lagrangian relaxation
joint optimization of incentive, caching, and routing policies based aggregated cost algorithm is employed to approximately
is proposed. solve the problem with lower complexity.
Chae and Choi [113] study the success rate of cached con- A distributed function computation in noisy multihop wire-
tent delivery in wireless caching helper networks, and draw less network scheme is considered in [118]. An adversarial
the conclusion that the success rate mainly depends on cache- noise model is adopted as the channel model, and only divisi-
based channel selection diversity and network interference. ble functions are considered. A general protocol for evaluating
They further point out that with given channel fading and any divisible functions is designed and the bottleneck of
network topology, both interference and channel selection the protocol is analyzed to offer insights of design of more
diversity vary corresponding to the caching strategies of the efficient protocols.
caching helpers. In order to maximize the average success You et al. [119] propose a solution for the integration
probability of content delivery, the paper studies probabilistic of mobile cloud computing and microwave power trans-
content placement and derives optimal caching probabilities fer (MPT), to offer computation services to passive low-
in closed form using stochastic geometry, to desirably control complexity devices such as sensors and wearable computing
channel selection diversity and network interference. devices. Mobile users can offload their computation through
BSs to cloud or execute them locally, and meanwhile, the
BSs can transfer power to mobile users. A set of policies for
B. Computing-Networking computation offloading decision and MPT time division are
1) Cloud Computing & Networking: Mobile cloud com- proposed, aiming at minimizing energy consumption and max-
puting (MCC) is a very typical paradigm of the integration of imizing harvested energy. The problem is solved by convex
computing and networking functionalities. By allowing mobile optimization theory.
users to offload their computation tasks to remote cloud, Zhang et al. [94] focus on the energy saving in mobile
which is typically located in the Internet, MCC endows exces- cloud computing systems with stochastic wireless channel. In
sive computational capabilities to wireless communication local computing case, the CPU frequency is reconfigured, and
networks, enabling the execution of computationally intensive in offloading case, the data transmission rate is dynamically
mobile applications such as natural language processing and adjusted according to the stochastic channel condition, both
augmented reality [1]. aiming at minimizing the mobile device energy consumption.
In order to leverage the intensive computation resource of The scheduling policies are formulated as constrained opti-
mobile cloud computing (MCC) on the processing of data mization problems, and closed-form solutions are obtained for
gathered by ubiquitous wireless sensor networks (WSNs), the problems.
some researches focus on the integration of these two network In [34], a solution for jointly considering wireless
paradigms. In [114], a framework concerning the processing networking and cloud computing is proposed to improve
of the sensory data in WSN–MCC integration is proposed. the end-to-end performances of cloud mobile media delivery.
The spectral efficiency in wireless network and pricing infor- as infrastructures to collaboratively provide computation and
mation in the cloud are formulated as a Stackelberg game communication services to the whole network.
model, in which the power allocation and interference manage- Jalali et al. [126] study the energy consumption issue of
ment are performed on the basis of pricing of media services. nano data centers (nDCs) in fog computing systems. Nano
The replicator dynamics method is adopted to solve this game. data centers are tiny computers that are distributed in end user
The work in [120] proposes a resource allocation scheme premises, and they serve as Internet of Things servers to han-
in cognitive wireless networks with cloud computing. A vir- dle content and applications in a peer-to-peer (P2P) fashion.
tualized data center model is derived to handle the dynamic Flow-based and time-based models are employed to investi-
resource allocation problem under fluctuating channel and gate the energy consumption of shared and unshared network
workload conditions, aiming at minimizing the energy con- equipment, respectively.
sumption and maximizing long-term average throughput while In order to provide sufficient computation resources for
maintaining network stability. video or image data processing in incidental disasters, the work
Liu et al. [121] consider wireless resource scheduling in of [58] proposes an incident-supporting visual cloud comput-
multiuser multiservice mobile cloud computing systems with ing solution, bringing computation resources from cloud to
different priorities for real-time services and non-real-time ser- the proximity of the incident sites. The proposed collection,
vices. An M/M/1 queueing model is formulated to improve computation and consumption solution leverages SDN to con-
the quality-of-service (QoS) of real-time streams and mini- struct a fog-cloud computing integrated architecture, in which
mize the average queueing delay of non-real-time services. fog computing is deployed at the network edge, close to data
A queueing delay-optimal control algorithm and a delay- collection/consumption sites, and coupling with cloud com-
constrained control algorithm are developed respectively to puting core, which provides major computation resource via
solve the problems. computation offloading operation.
Chiang et al. [122] investigate the security issues in a virtu- Zeng et al. [127] study the resource management and task
alized cloud computing paradigm–Amazon Elastic Compute scheduling issues in fog computing supported software-defined
Cloud (EC2). They demonstrate that in EC2, an adver- embedded system, in which computation tasks can be executed
sary can deliberately delay the execution of targeted appli- either by embedded devices or by a fog computing server. The
cation, by utilizing the competition for shared resources resource allocation, the I/O interrupt requests balance issues
between virtual I/O workloads. A framework called Swiper and the task scheduling between mobile devices and fog com-
is designed and presented as a tool to slow down the tar- puting servers are formulated into one mixed-integer nonlinear
geted application and virtual machine with minor resource programming problem to search for effective solutions.
consumption. In order to manage the physical diversity of devices and
2) Fog Computing & Networking: In order to reduce the bridge different networks, [128] proposes a type of special
transmission latency of cloud computing systems and provide fog node, called IoT Hub. A IoT Hub is placed at the edge
localized and location-based information services, fog comput- of multiple networks, offering the services of border router,
ing is proposed as a complement of cloud computing, bringing cross-proxy, cache, and resource directory.
computation resources into proximity to end users. In essence, Due to the distributed nature of fog nodes and uncertainty on
fog computing serves as an intermediate layer between the arrival of fog nodes within the network, the dynamic and opti-
cloud and mobile devices. mal fog network formation and task distribution is a problem
Deng et al. [123] study the cooperation and interplay that needs to be solved. The work in [129] is dedicated to this
between the fog and the cloud. The workload allocation topic. An approach based on the online secretary framework is
between the fog and the cloud is formulated as an optimiza- proposed. In the framework, any given task initiating fog node
tion problem, aiming at minimizing power consumption under is able to choose its desired set of neighboring fog nodes and
the service delay constraint. An approximate algorithm which perform task offloading, aiming at minimizing the maximum
decomposes the primal problem into several subproblems is computational latency.
hired to solve the original problem. 3) Mobile Edge Computing & Networking: As a recently
In [124], a new framework called device-to-device (D2D) emerging computing-networking paradigm, mobile edge com-
fogging is proposed to enable the sharing of computation and puting (MEC) aims at endowing the wireless communication
communication resources among mobile devices. The frame- networks with low-latency computation services, by allow-
work is enabled by D2D collaboration under the assistance ing mobile users to offload their computationally intensive
of network control. An optimization problem is formulated workloads to MEC servers placed in proximity to end users.
in this framework to minimize the time-average energy con- The major issues in MEC systems are computation offloading
sumption of task execution of all the mobile devices, taking decision and resources allocation.
into consideration the incentive constraints to prevent the over- Reference [63] simply focuses on the offloading decision
exploiting and free-riding behaviors. An online task offloading problem in MEC systems. The investigation of the paper
algorithm that leverages Lyapunov optimization methods is argues that the multi-user computation offloading problem in
derived accordingly. a multi-channel wireless interference environment is an NP-
Similar to [124], the work in [125] presents a distributed hard problem, so the problem is formulated into a multi-user
architecture in vehicular communication network, called vehic- computation offloading game, and further investigation on the
ular fog computing, in which vehicle end users are utilized game property shows that the game admits a Nash equilibrium.
Therefore, a distributed offloading algorithm that achieves a Nowadays, large-scale video distribution constitutes a con-
Nash equilibrium is designed accordingly to solve the game. siderably large proportion of the Internet traffic. In [134], a
In order to reduce the frequency of interrupted computation solution utilizing Media Cloud is proposed to deliver on-demand
tasks due to battery depletion and achieve green comput- adaptive video streaming services. The caching, transcoding
ing, [99] investigates a green MEC system with energy and bandwidth costs at each edge server are formulated as an
harvesting technology and proposes a computation offloading optimization problem, aiming at minimizing the total opera-
and resource allocation solution in this system, taking into tional cost. Subsequently, the closed-form solution of optimal
account the offloading decision, the CPU-cycle frequencies caching space allocation and transcoding configuration at each
for mobile execution, and the transmitting power for com- edge server is derived through a two-step approach.
putation offloading. Aiming at minimizing the execution cost, As a typical application of integrated caching, comput-
which addresses execution latency and task failure, a Lyapunov ing and networking systems, video transcoding and caching
optimization-based dynamic computation offloading algorithm received considerable attention in literature. Reference [135]
is presented for problem solving. proposes a partial transcoding scheme for content management
Reference [95] considers the joint optimization of radio in heterogenous networks with media cloud. Video content is
resource (the transmit precoding matrices of the mobile encoded and separated into segments, part of which are stored,
devices) and computation resource (the assigned CPU resulting in caching cost, while others are transcoded online,
cycles/second) during MEC operations in a MIMO multicell resulting in computing cost. In this context, a constrained
system. This issue is formulated as an optimization problem, stochastic optimization problem is formulated to determine
aiming at minimizing the overall users energy consumption, whether a segment is cached or transcoded, aiming at min-
while meeting latency constraints. The closed-form global imizing the long-term overall cost. An algorithm utilizing
optimal solution is derived in single-user case, and an iterative Lyapunov optimization framework and Lagrangian relaxation
algorithm utilizing successive convex approximation technique is designed to solve the problem.
is applied to solve the problem in multiuser scenario. Similar to [135], reference [136] considers the caching-
Unlike the studies in [63] and [99], the work in [130] computing tradeoff problem in cloud computing environment.
considers partial computation offloading approach, in which For each query, a Web application can choose either to com-
only part of the computation task of a UE is offloaded to pute the response fresh, which causes computing cost, or to
the MEC serve, while the rest of it is executed locally in cache the response to reduce future computing cost, which
device. Two optimization problems are designed with objec- causes caching cost. This problem is abstracted as a variant
tives of energy consumption of UE minimization (ECM) and of the classical Ski-Rental problem. A randomized algorithm
latency of application execution minimization (LM), respec- is proposed accordingly to provide arrivals-distribution-free
tively. The computational speed of UEs, transmit power of performance guarantees.
UEs and offloading ratio are adopted as variables in these A simulation framework is proposed in [137], to serve as
two optimization problems. The non-convex ECM problem is an evaluation tool for transparent computing systems with
transformed into a convex problem by using variable substitu- particular cache schemes. The simulation framework enables
tion technique and solved, and the LM problem is solved by a the evaluation of the performance of multi-level cache hierar-
locally optimal algorithm with the univariate search technique. chies with various cache replacement policies in transparent
We propose an integrated framework for interference man- computing systems.
agement and computation offloading in wireless networks with Reference [138] utilizes big data tool to investigate the
MEC in [131] and [132]. In the framework, the computation proactive caching issue in 5G wireless networks, aiming at
offloading decision, physical resource block allocation and optimizing the caching gain in terms of request satisfactions
MEC computation resource allocation are formulated as an and backhaul offloadings. This work proposes that content
optimization problem. The computation offloading decisions popularity estimation plays an important role in optimization
are made by the MEC server according to load estimations, of caching gain. The users mobile traffic data in the network
then the PRBs are allocated by a graph colouring method. is collected and analyzed by big data platforms deployed at
Finally, the outcomes of PRB allocation and offloading deci- base stations for content popularity estimation.
sions are used to distribute computation resources to UEs. Reference [139] focuses on the study of moving data-
intensive video distribution applications to Cache A Replica
On Modification (CAROM) cloud file system, aiming at reduc-
C. Caching-Computing-Networking ing the access latency while maintaining low storage cost.
Reference [133] considers data prefetching issue in wireless A scheduling mechanism is proposed to serve as a lubri-
computing over fluctuating channel conditions. The mobile cant between data-intensive applications and CAROM. In
users can prefetch data for computing from access points when the proposed scheme, a tripartite graph is adopted to clarify
the channel condition is good, in anticipation of requesting the relationships among computation nodes, data nodes and
these data during bad quality channel periods. The issue is tasks. A framework is designed for the situation in which the
whether to prefetch or not in each time point, taking into con- task data is stored in the cache, followed by the design of
sideration memory constraints, application latency and channel two variant frameworks considering the performance of task
fluctuations. A dynamic programming approach is designed and limitations of cache size. Finally, a k-list algorithm is
accordingly to derive the optimal prefetching policy. employed to serve as an approximation algorithm.
As an efficient tool for resource sharing, network vir- 1) Stringent Latency Requirements: One of the foreseeable
tualization has been widely applied to cloud computing application scenarios of the integrated networking, caching
systems. Virtualization approaches can consolidate applica- and computing system is Internet of Things (IoT) [144]. A
tion workloads into shared servers. Due to multiple computing lot of industrial control systems, such as oil and gas systems,
cores, multi-level caching hierarchy and memory subsys- goods packaging systems, and manufacturing systems, typi-
tems, there exist contentions between virtualized resources. cally require the end-to-end latency from sensors to control
Reference [140] proposes the hardware platform specific coun- nodes be bounded within a few milliseconds [145]–[147].
ters to detect those contentions. Furthermore, a software probe Other potential applications of integrated networking, caching
based scheme is designed to enhance the contention detection. and computing system like virtual reality applications, drone
Reference [141] studies the decentralized coded content flight control systems, and vehicle-to-roadside communica-
caching in multihop device-to-device communication scenario tions, often require latency to be under a few tens of millisec-
of next generation cellular networks. The contents are linearly onds [144]. These stringent latency requirements fall outside
combined and stored in caches of user devices and femtocells, of the capability of current systems, and therefore pose a
and the throughput capacity is compared to that of decentral- great challenge for the integrated networking, caching and
ized uncoded content caching. It is demonstrated that, due computing system.
to content coding, the number of hops needed to deliver the 2) Tremendous Amount of Data Against Network
requested content is reduced, and thereby the decentralized Bandwidth Constraints: As an increasing number of
coded content caching can achieve a better network through- mobile devices are connecting into the wireless networks,
put capacity compared to decentralized uncoded caching. a tremendous and exponentially growing amount of data is
Moreover, it is shown that under Zipfian content request dis- being created by them [148]. For instance, the data generated
tribution, the decentralized coded content cache placement is by an autonomous vehicle is estimated to be about one
able to increase the throughput of cellular networks by a factor gigabyte per second [149]. The data created by the U.S.
of (log(n))2 , where n stands for the number of nodes served smart grid is expected to be 1000 petabytes each year [150],
by a femtocache. and the U.S. Library of Congress can generate data of
We formulate the computation offloading decision, con- 2.4 petabytes per month. By comparison, Google traffics
tent caching strategy and spectrum and computation resources around one petabyte data one month. Sending all the data
allocation issues as an optimization problem in our previous to the edge, the fog or the cloud may require extremely
papers [142], [143]. The original non-convex problem is high network bandwidth, and in the mean time, cause
transformed into a convex problem and further decomposed prohibitively high costs. Furthermore, due to data privacy
into a number of sub-problems. Then a distributed algorithm concerns or regulations, sometimes the data should not be
called alternating direction method of multipliers (ADMM) disseminated. Therefore, the central issue here is the design
is employed to solve the problem in a more efficient way of architectures of integrated systems in which the data can
compared to a centralized algorithm. be absorbed (stored) and digested (processed) locally, without
Finally, the enabling technologies discussed above are sum- long distance propagations.
marized in Table III and Table IV. Furthermore, we also 3) Uninterruptable Services Against Intermittent
provide a comparison of pros and cons of all the technologies Connectivity: Some of the devices in the network may
in Table V and Table VI. suffer from widely fluctuating quality of wireless channels
and intermittent connectivity to edge/fog/cloud servers. Such
VII. C HALLENGES AND B ROADER P ERSPECTIVES devices may include vehicles, drones and sometimes mobile
terminals of cellular communication systems. However,
Despite the promising prospect of the integration of
applications and services like data collection, data analytics
networking, caching and computing, a number of significant
and controls need to be available at all times. Therefore,
research challenges remain to be addressed prior to widespread
novel system architectures, service time schedulings, caching
implementation of integrated networking, caching and comput-
mechanisms and resource allocation schemes that guarantee
ing systems. In this section, we present some of these research
consistent availability of services under fluctuating channel
challenges, followed by a discussion on broader perspectives.
conditions and intermittent connectivity to servers are in
urgent demand.
A. Challenges 4) Interference of Multiple Interfaces: In order to pro-
The integrated networking, caching and computing design vide resources to all types of end users, as well as uti-
paradigm raises opportunities as well as challenges in both lize the caching and computing resources of end users,
academia and industry. On academic side, the classical multiple interfaces (e.g., cellular, WiFi, WiMAX) should be
information theoretic models that anchored numerous break- valid in the integrated systems. All the users should be
throughs in traditional telecommunication technologies cannot able to access the same functionalities through a variety of
be simply applied to integrated networking, caching and com- interfaces. However, the communication efficiency may be sig-
puting systems. Similarly on the industrial side, the classical nificantly undermined by co-existence of multiple interfaces
protocols designed for early stages of wireless communication due to paucity of channels. Current technologies that have
systems are inadequate for leveraging integrated resources on the potential to enable simultaneous communications via
three dimensions. multiple interfaces like SDN, MIMO and cognitive radio in
TABLE III
E NABLING T ECHNOLOGIES FOR I NTEGRATED N ETWORKING , C ACHING AND C OMPUTING
general require multiple access techniques, which are cur- 5) Network Effectiveness in the Face of Mobility: In
rently not scalable for massive multiple interfaces wireless the integrated systems, a considerable number of content
networks [151]. providers, caching nodes and computing nodes may be mobile
TABLE IV
E NABLING T ECHNOLOGIES FOR I NTEGRATED N ETWORKING , C ACHING AND C OMPUTING (C ONTINUED )
devices [152]. In this context, the central issue is how the can utilize both µW and mmW frequencies. Fundamental
network can know where to find these caching and com- results on the caching capabilities and average achievable rate
puting resources providing nodes. Data request failures and of caching are derived, and the impact of caching on average
route failures may be increased by this nodes mobility, and handover failure and energy consumption is analyzed.
hence deteriorate the overall system performance. Therefore, However, pertinent research works only offer a prelim-
addressing the mobility issue is imperative for the guarantee inary perspective and raise a number of open issues. For
of network effectiveness. instance, 1-dimensional mobility models are usually employed
On the other hand, the integrated system could benefit from in analysis, and one single destination is typically considered,
the prediction of user mobility. ICN could develop more effi- despite the fact that the users can pass by multiple destina-
cient caching strategies and place the requested content close tions in practice. Furthermore, the decision on what services
to users’ predicted locations in advance, provided that accu- should be migrated and the question whether group mobility
rate and timely prediction of user mobility is performed [153]. is better than single user mobility, have not been adequately
Moreover, provided with user and group mobility predictions, addressed [153].
systems like Mobile Follow-me Cloud (M-FMC) could achieve 6) The Networking-Caching-Computing Capacity: Deter-
optimal content migration at different cache levels according mining the theoretical capacity of each dimension of the
to the proximity of users, fulfilling the goals of MEC [153]. integrated system is a critical task that remains to be addressed.
Reference [154] takes the user mobility into account Since the classical information theoretical model concern-
and proposes a proactive caching approach for named data ing instantaneous rate regions cannot sufficiently address the
networking (NDN). In this approach, the content requested but caching-induced non-causality within the system, it cannot
not received by a mobile user before a handover is proactively be directly applied on the integrated systems. The study
cached, and is delivered to that user right after the handover. in [156] shows that the traditional concept of rate capac-
As such, the content delivery ratio can be increased. ity is inadequate to represent the network capability of
In order to analyze and manage mobility in joint microwave delivering non-private contents for multiple users. In order
(µW)–millimeter wave (mmW) networks, [155] proposes an to reflect the impact of caching on content delivery, this
approach that leverages device-level caching along with dual- work proposes a so-called content rate to measure the rate
mode small base stations. The dual-mode small base stations at which the cached content is delivered to users using
TABLE V
A DVANTAGES AND S HORTCOMINGS OF THE E NABLING T ECHNOLOGIES
shared channels. Despite the plethora of studies on the incor- calculation remains ambiguous. One category of approaches
poration of computing into wireless networks, the role of concerning this matter is networking coding, in which the
computing in the integrated system capacity measurement and algebraic operation (i.e., computing) is distinguished from
TABLE VI
A DVANTAGES AND S HORTCOMINGS OF THE E NABLING T ECHNOLOGIES (C ONTINUED )
the communication operation [157]. Other approaches like sizes, these distributed infrastructures may not have suffi-
distributed MIMO [158] and base station collaborative trans- cient resource to detect threats and protect themselves from
mission [151], do not decouple computing from communica- attacks.
tion, as channel coding and logic operations are intertwined. 9) Convergence and Consistency: A considerable number
Needless to say, a unified capacity analysis in which the of integrated networking, caching and computing systems are
networking, caching and computing resources are represented organized in distributed manners, which may cause inconsis-
in canonical form (as in Fig. 2), bears theoretical significance tency, oscillation and divergence to the global system states,
and is therefore in urgent demand. and this may become especially acute in the case of mas-
7) The Networking-Caching-Computing Tradeoffs: The sive, under-organized mobile system with a virtualized pool
study in [3] demonstrates that the same types of services can of unpredictably shared resources. The convergence and con-
be provided by the integrated system utilizing different com- sistency issue is a typical and recurrent challenge in distributed
binations of networking, caching and computing resources. systems, and some of the use cases of integrated systems
However, the tradeoff among networking, caching and com- like stream mining and edge analytics have posed exceptional
puting resources for each service type, should be analyzed demand on addressing this challenge.
individually against the associated resource constraints and 10) End-to-End Architectural Tradeoffs: This new
performance measures. In this context, the central issue lies paradigm of system design offers new opportunities of
on the optimal tradeoff of resources for each individual case, creating better tradeoffs between centralized and distributed
which is clearly nontrivial to determine. architectures, between deployment planning and resilience
8) Security: The new architectures of the integrated via redundancy, and between what should stay local and
networking, caching and computing systems present new secu- what should go global. Moreover, logical integrated system
rity challenges. Due to the fact that cloud servers are typically topologies, either dynamically or statically established, over
installed in heavily protected facilities selected and controlled the same underlying physical infrastructures can facilitate the
by cloud operators, cloud systems are relatively secure against realization of a spectrum of system frameworks from fully
attacks. However, in order to better meet the user require- centralized to completely distributed.
ments, fog/edge servers and caching facilities are usually
deployed in distributed manner and in proximity to end users– B. Broader Perspectives
where users need them to be. Such distributed systems are Since the integrated networking, caching and computing
generally vulnerable to attacks. Moreover, due to limited system is currently in its infancy, its development course can
be influenced by a lot of other technologies. In the mean time, Though there are many differences between each other, they
this integrated architecture may impact them as well. Here we are actually highly complementary technologies that can work
briefly discuss these technologies and the potential interac- together. SDN can help NFV in addressing the challenges of
tions and benefits when applying these technologies into the dynamic resource management and intelligent function orches-
integrated system. tration. Through the implementation of NFV, SDN is able to
1) Software-Defined Networking: Software defined create a virtual service environment dynamically for a specific
networking (SDN) endows programmability to the networks type of service chain [161].
by introducing a logically software-defined controller and With the help of SDN and NFV, the integrated system can
consequently separating the control plane from the data realize a dynamic and efficient orchestration of networking,
plane [18], [159]. By making network configuration decisions caching and computing resources to improve the overall
in a centralized controller that has a global view of the performance of the entire system.
network, SDN enables logical centralization of control [5]. 3) Wireless Network Virtualization: Motivated by the
Due to its intrinsic virtues, SDN can easily introduce new success of virtualization in wired networks, the concept
abstractions in networking, simplify network management, of network virtualization has been extended to wireless
enable dynamic network reconfiguration and facilitate network networks [14]. Wireless network virtualization technology
evolution. Furthermore, SDN can efficiently manage and enables the decouple of wireless network infrastructure from
optimize network resource allocation in response to time- the services that it provides, so that differentiated services can
varying network conditions [160]. Due to the effectiveness share the same network infrastructure, thereby maximizing
and efficiency of SDN on managing the wireless network its utilization [163]. Wireless network virtualization has the
systems, it is natural to employ SDN as the control unit of potential to efficiently share and rationally utilize the physi-
the integrated networking, caching and computing system. cal network resources, to rapidly respond to dynamic network
However, it is worth noting that the control and resource changes, and to innovatively realize multiple customized ser-
allocation mechanism of the integrated system is a much vices. Furthermore, by isolating part of the network, wireless
broader concept than SDN, and its diversity should go far network virtualization endows the network with capability of
beyond any single technology. easily migrating to up-to-date technologies while preserving
2) Network Function Virtualization: Network function vir- legacy technologies [22]. It should be noted that the sharing
tualization (NFV) realizes network functions as pure software of resources enabled by wireless network virtualization is not
on commodity and general hardwares, which can solve the limited to sharing of networking infrastructures like BSs and
problem of network ossification, and reduce the CapEx and routers, but to the extent of sharing of cached content and
OpEx [161]. NFV is network virtualization on steroids. Instead unoccupied computational resources. That is to say, not only
of controlling the network resources, NFV controls network the networking but also the caching and computing resources
functions such as firewall, load balancer, network address can be efficiently shared by end users from different virtual
translation, intrusion detection, etc. Network virtualization networks through wireless network virtualization. As a conse-
makes it possible to create multiple virtual networks on top of quence, the CapEx and OpEx of wireless access networks and
the same physical network. By contrast, NFV enables creating core networks, can be significantly reduced. If network virtu-
network functions on top of general-purpose servers instead of alization were applied to the integrated networking, caching
the traditional middleboxes. and computing system, both the infrastructure and the con-
Recently, there is an increasing trend of integrating tent within the system can be shared among multiple virtual
NFV with SDN to jointly optimize network functions and networks, which exhibits great potential to exploit the three
resources [161], [162]. SDN and NFV are two closely related dimensions of resources/capacities to the hilt. However, this
technologies. Both SDN and NFV move toward network vir- in turn increases the complexity of the system, hence raising
tualization and automation, however, they have different con- new requirements on resiliency, consistency, scalability and
cepts, system architectures, and functions that are summarized convergence, as well as resource allocation and management
as follows. approaches of the entire system.
• SDN is a concept of achieving centrally controlled and 4) Big Data Analytics: Joint consideration of networking,
programmable network architecture to provide better con- caching and computing will definitely increase the complex-
nectivity. By contrast, NFV is a concept of implementing ity of the system, and raise new requirements on resource
network functions in a software manner. allocation and management approaches. In the meantime, the
• SDN aims at providing rapid application deployment and integrated system will inevitably generate enormous amount
service delivery through abstracting the underlying infras- of data, such as the big signaling data, and big traffic data.
tructure. By contrast, NFV aims at reducing the CapEx, We will explain more about these two kinds of illustrative big
OpEx, space, and power consumption. data in the following.
• SDN decouples the network control and data forward- a) Big signaling data: Control message plays an impor-
ing functions, which enables the directly programmable tant role in the part of networking for mobile users, and
network central control. By contrast, NFV decouples the the mutual communications between networking, caching and
network functions from the dedicated hardware devices, computing functionalities. The control message is also termed
which provisions flexible network functionalities and as signaling data. The signaling data works on the basis of
deployment. predefined protocols to guarantee wireless communication’s
security, reliability, regularity and efficiency. In the integrated and computing, the system complexity could be very high.
system of networking, caching and computing, there must Deep reinforcement learning is a recently emerging tech-
be large amount of signaling data, such as the signaling nology, which integrates deep learning with reinforcement
interaction procedure of the networking-caching framework as learning algorithm to handle large amount of input data and
illustrated in Fig. 4. obtain the best policy for tough problems. Deep reinforce-
Signaling monitoring system is a significantly important ment learning utilizes deep Q-network to approximate the Q
support system for the operation and maintenance of wire- value function [167]. Google Deepmind adopts this method
less communications. It analyzes the rich signaling data and on some games [167], [168], and gets quite good results.
assessment indicators to realize real-time network monitor- Furthermore, deep reinforcement learning has been explored
ing, handle customer complaints, troubleshoot the network, in various wireless networks [169]. For the integrated system
block illegal traffic, improve the network quality, etc. With the of networking, caching, and computing, deep reinforcement
explosive growth of mobile users, and the increasing demand learning can be used as a powerful tool to obtain good
for high-speed data, the amount and types of the signaling resource allocation policies [170].
data will rise up tremendously in the integrated system of
networking, caching and computing. Therefore the traditional VIII. C ONCLUSION
signaling monitoring systems may be faced with many tough
This article addressed the integration of networking, caching
problems, and some insightful information is unable to be dis-
and computing, which is becoming an important concept that
covered. Big data analytics is a recently proposed powerful
leverages the three types of resources to meet the diversity
tool to help analyze the big signaling data [164]. For example,
and intensiveness of user requirements on mobile services.
Celebi et al. [165] analyze the base station system applica-
We began our discussion with an overview of networking,
tion part (BSSAP) messages from A interface in a Hadoop
caching and computing in wireless systems, in which the
platform to identify handovers from 3G to 2G, and the sim-
brief history and recent advances of these three aspects
ulation results indicate that the identified 3G coverage holes
were succinctly introduced. We then discussed the moti-
are in agreement with the drive test conclusions. This exam-
vations and requirements of the integration of networking,
ple serves as an inspiration of leveraging big data analytics in
caching and computing. Next, we discussed the frame-
wireless networks, and it should be promising to incorporate
works of the integrated system in four categories, namely,
the big data analytics into the integrated networking, caching
caching-networking framework, computing-networking frame-
and computing framework to further optimize the system.
work, caching-computing framework and caching-computing-
b) Big traffic data: With the development of wireless
networking framework. We then discussed some performance
technologies and the prosperity of social applications, users
metrics that can be used to compare different architectures,
upload and download lots of video-related content, which
management systems, resource allocation algorithms, cus-
results in vast amount of traffic data that keeps increasing
tomization flexibility, interfaces, energy saving mechanisms,
at an amazing rate. Traffic data is continuously generated and
etc. Next, some enabling technologies for the integration of
carried on in the integrated networking, caching and comput-
networking, caching and computing were discussed according
ing system, and the operators have to work hard on balancing
to types of integration. We also discussed some signifi-
the network load and optimizing the network utilization.
cant research challenges in the integrated system, including
Traffic monitoring and analyzing is an essential component
latency requirement, bandwidth constraints, interfaces, mobil-
in network management, which enables the maintenance of
ity management, security, resource and architecture tradeoffs,
smooth network operations, failure detection and prediction,
convergence, etc. Finally, some broader perspectives were
security management, etc. Traditional approaches to monitor-
explored.
ing and analyzing the traffic data are impotent when faced
In summary, research on the integrated networking, caching
with the big traffic data. SDN can be combined with big data
and computing systems is quite broad and a number of chal-
analytics, which would bring some excellent advantages [21].
lenges lay ahead. Nevertheless, it is in favor of the wireless
The topic of traffic characteristics has attracted a lot of atten-
community to swiftly address the challenges and go forward.
tion, due to the fact that understanding the traffic dynamics
This article attempts to briefly explore the technologies related
and usage conditions is of great importance for improving the
to the integration of networking, caching and computing at
network performance. For example, Liu et al. [166] use big
a very preliminary level, and to discuss future research that
data analytics to investigate three features of the network traf-
may benefit the pursuit of this vision. We hope that our dis-
fic: network access time, traffic volume, and diurnal patterns.
cussion and exploration here may open a new avenue for the
The results reveal some insightful information and provide
development of integrated networking, caching and computing
a reference for the operators to take proactive actions for
systems, and shed faint lights on effective and efficient designs
network capacity management and revenue growth. It would
therein.
be beneficial to introduce the big data analytics into the
integrated networking, caching and computing system to facil-
itate traffic monitoring and network optimization, and acquire ACKNOWLEDGMENT
bright insights therein. The authors thank the reviewers for their detailed reviews
5) Deep Reinforcement Learning: When we jointly and constructive comments, which have helped to greatly
consider the three technologies: networking, caching improve the quality of this paper.
R EFERENCES [24] R. Xie, F. R. Yu, H. Ji, and Y. Li, “Energy-efficient resource alloca-
tion for heterogeneous cognitive radio networks with femtocells,” IEEE
[1] S. Wang and S. Dey, “Adaptive mobile cloud computing to enable Trans. Wireless Commun., vol. 11, no. 11, pp. 3910–3920, Nov. 2012.
rich mobile multimedia applications,” IEEE Trans. Multimedia, vol. 15, [25] M. Agiwal, A. Roy, and N. Saxena, “Next generation 5G wireless
no. 4, pp. 870–883, Jun. 2013. networks: A comprehensive survey,” IEEE Commun. Surveys Tuts.,
[2] S. Abolfazli, Z. Sanaei, E. Ahmed, A. Gani, and R. Buyya, “Cloud- vol. 18, no. 3, pp. 1617–1655, 3rd Quart., 2016.
based augmentation for mobile devices: Motivation, taxonomies, and [26] T. O. Olwal, K. Djouani, and A. M. Kurien, “A survey of resource man-
open challenges,” IEEE Commun. Surveys Tuts., vol. 16, no. 1, agement toward 5G radio access networks,” IEEE Commun. Surveys
pp. 337–368, 1st Quart., 2014. Tuts., vol. 18, no. 3, pp. 1656–1686, 3rd Quart., 2016.
[3] H. Liu, Z. Chen, and L. Qian, “The three primary colors of mobile [27] L. Ma, F. Yu, V. C. M. Leung, and T. Randhawa, “A new method to
systems,” IEEE Commun. Mag., vol. 54, no. 9, pp. 15–21, Sep. 2016. support UMTS/WLAN vertical handover using SCTP,” IEEE Wireless
[4] Y. He, F. R. Yu, N. Zhao, V. C. M. Leung, and H. Yin, “Software- Commun., vol. 11, no. 4, pp. 44–51, Aug. 2004.
defined networks with mobile edge computing and caching for smart [28] F. Yu and V. Krishnamurthy, “Optimal joint session admission con-
cities: A big data deep reinforcement learning approach,” IEEE trol in integrated WLAN and CDMA cellular networks with vertical
Commun. Mag., vol. 55, no. 12, Dec. 2017. handoff,” IEEE Trans. Mobile Comput., vol. 6, no. 1, pp. 126–139,
[5] R. Huo et al., “Software defined networking, caching, and computing Jan. 2007.
for green wireless networks,” IEEE Commun. Mag., vol. 54, no. 11, [29] S. Bu and F. R. Yu, “Green cognitive mobile networks with small cells
pp. 185–193, Nov. 2016. for multimedia communications in the smart grid environment,” IEEE
[6] Q. Chen et al., “An integrated framework for software defined Trans. Veh. Technol., vol. 63, no. 5, pp. 2115–2126, Jun. 2014.
networking, caching, and computing,” IEEE Netw., vol. 31, no. 3, [30] W. Cheng, X. Zhang, and H. Zhang, “Statistical-QoS driven energy-
pp. 46–55, May/Jun. 2017. efficiency optimization over green 5G mobile wireless networks,” IEEE
[7] C. Fang, F. R. Yu, T. Huang, J. Liu, and Y. Liu, “A survey of green J. Sel. Areas Commun., vol. 34, no. 12, pp. 3092–3107, Dec. 2016.
information-centric networking: Research issues and challenges,” IEEE [31] G. Huang and J. Li, “Interference mitigation for femtocell networks via
Commun. Surveys Tuts., vol. 17, no. 3, pp. 1455–1472, 3rd Quart., adaptive frequency reuse,” IEEE Trans. Veh. Technol., vol. 65, no. 4,
2015. pp. 2413–2423, Apr. 2016.
[8] C. Liang, F. Yu, and X. Zhang, “Information-centric network function [32] A. Checko et al., “Cloud RAN for mobile networks—A technology
virtualization over 5G mobile wireless networks,” IEEE Netw., vol. 29, overview,” IEEE Commun. Surveys Tuts., vol. 17, no. 1, pp. 405–426,
no. 3, pp. 68–74, May/Jun. 2015. 1st Quart., 2015.
[9] Z. Sanaei, S. Abolfazli, A. Gani, and R. Buyya, “Heterogeneity [33] Y. Cai, F. R. Yu, and S. Bu, “Cloud computing meets mobile wireless
in mobile cloud computing: Taxonomy and open challenges,” IEEE communications in next generation cellular networks,” IEEE Netw.,
Commun. Surveys Tuts., vol. 16, no. 1, pp. 369–392, 1st Quart., 2014. vol. 28, no. 6, pp. 54–59, Nov./Dec. 2014.
[10] S. Andreev et al., “Exploring synergy between communications, [34] Z. Yin, F. R. Yu, S. Bu, and Z. Han, “Joint cloud and wireless
caching, and computing in 5G-grade deployments,” IEEE Commun. networks operations in mobile cloud computing environments with tele-
Mag., vol. 54, no. 8, pp. 60–69, Aug. 2016. com operator cloud,” IEEE Trans. Wireless Commun., vol. 14, no. 7,
[11] M. Richart, J. Baliosian, J. Serrat, and J.-L. Gorricho, “Resource slicing pp. 4020–4033, Jul. 2015.
in virtual wireless networks: A survey,” IEEE Trans. Netw. Service [35] G. Fodor et al., “An overview of massive MIMO technology compo-
Manag., vol. 13, no. 3, pp. 462–476, Sep. 2016. nents in METIS,” IEEE Commun. Mag., vol. 55, no. 6, pp. 155–161,
[12] X. Zhou, R. Li, T. Chen, and H. Zhang, “Network slicing as a service: Jul. 2017.
Enabling enterprises’ own software-defined cellular networks,” IEEE [36] K. N. R. S. V. Prasad, E. Hossain, and V. K. Bhargava, “Energy
Commun. Mag., vol. 54, no. 7, pp. 146–153, Jul. 2016. efficiency in massive MIMO-based 5G networks: Opportunities and
[13] N. Bizanis and F. A. Kuipers, “SDN and virtualization solutions for challenges,” IEEE Wireless Commun., vol. 24, no. 3, pp. 86–94,
the Internet of Things: A survey,” IEEE Access, vol. 4, pp. 5591–5606, Jan. 2017.
2016. [37] L. You, X. Gao, G. Y. Li, X.-G. Xia, and N. Ma, “BDMA for
[14] C. Liang and F. R. Yu, “Wireless network virtualization: A survey, millimeter-wave/terahertz massive MIMO transmission with per-beam
some research issues and challenges,” IEEE Commun. Surveys Tuts., synchronization,” IEEE J. Sel. Areas Commun., vol. 35, no. 7,
vol. 17, no. 1, pp. 358–380, 1st Quart., 2014. pp. 1550–1563, Jul. 2017.
[15] C.-F. Lai, R.-H. Hwang, H.-C. Chao, M. M. Hassan, and A. Alamri, [38] L. Kong, M. Khan, F. Wu, G. Chen, and P. Zeng, “Millimeter-wave
“A buffer-aware HTTP live streaming approach for SDN-enabled wireless communications for IoT-cloud supported autonomous vehi-
5G wireless networks,” IEEE Netw., vol. 29, no. 1, pp. 49–55, cles: Overview, design, and challenges,” IEEE Commun. Mag., vol. 55,
Jan./Feb. 2015. no. 1, pp. 62–68, Jan. 2017.
[16] P. K. Agyapong, M. Iwamura, D. Staehle, W. Kiess, and A. Benjebbour, [39] S. Han, Chih-Lin I, Z. Xu, and C. Rowell, “Large-scale antenna systems
“Design considerations for a 5G network architecture,” IEEE Commun. with hybrid analog and digital beamforming for millimeter wave 5G,”
Mag., vol. 52, no. 11, pp. 65–75, Nov. 2014. IEEE Commun. Mag., vol. 53, no. 1, pp. 186–194, Jan. 2015.
[17] Q. Yan, F. R. Yu, Q. Gong, and J. Li, “Software-defined [40] N. Ishikawa, R. Rajashekar, S. Sugiura, and L. Hanzo, “Generalized-
networking (SDN) and distributed denial of service (DDoS) attacks spatial-modulation-based reduced-RF-chain millimeter-wave commu-
in cloud computing environments: A survey, some research issues, and nications,” IEEE Trans. Veh. Technol., vol. 66, no. 1, pp. 879–883,
challenges,” IEEE Commun. Surveys Tuts., vol. 18, no. 1, pp. 602–622, Jan. 2017.
1st Quart., 2016. [41] R. Ma, N. Xia, H.-H. Chen, C.-C. Chiu, and C.-S. Yang, “Mode selec-
[18] D. Kreutz et al., “Software-defined networking: A comprehensive tion, radio resource allocation, and power coordination in D2D com-
survey,” Proc. IEEE, vol. 103, no. 1, pp. 14–76, Jan. 2015. munications,” IEEE Wireless Commun., vol. 24, no. 3, pp. 112–121,
[19] N. McKeown et al., “OpenFlow: Enabling innovation in campus Feb. 2017.
networks,” SIGCOMM Comput. Commun. Rev., vol. 38, no. 2, [42] Y. Cai, F. R. Yu, C. Liang, B. Sun, and Q. Yan, “Software-defined
pp. 69–74, Apr. 2008. device-to-device (D2D) communications in virtual wireless networks
[20] J. Liu, S. Zhang, N. Kato, H. Ujikawa, and K. Suzuki, “Device-to- with imperfect network state information (NSI),” IEEE Trans. Veh.
device communications for enhancing quality of experience in software Technol., vol. 65, no. 9, pp. 7349–7360, Sep. 2016.
defined multi-tier LTE-A networks,” IEEE Netw., vol. 29, no. 4, [43] K. Wang, F. R. Yu, H. Li, and Z. Li, “Information-centric wireless
pp. 46–52, Jul./Aug. 2015. networks with virtualization and D2D communications,” IEEE Wireless
[21] L. Cui, F. R. Yu, and Q. Yan, “When big data meets software-defined Commun., vol. 24, no. 3, pp. 104–111, Jan. 2017.
networking: SDN for big data and big data for SDN,” IEEE Netw., [44] Y. Zhou and W. Yu, “Optimized backhaul compression for uplink cloud
vol. 30, no. 1, pp. 58–65, Jan./Feb. 2016. radio access network,” IEEE J. Sel. Areas Commun., vol. 32, no. 6,
[22] C. Liang and F. R. Yu, “Virtual resource allocation in information- pp. 1295–1307, Jun. 2014.
centric wireless virtual networks,” in Proc. IEEE Int. Conf. [45] A. Ioannou and S. Weber, “A survey of caching policies and forward-
Commun. (ICC), London, U.K., Jun. 2015, pp. 3915–3920. ing mechanisms in information-centric networking,” IEEE Commun.
[23] J. Hoadley and P. Maveddat, “Enabling small cell deployment with Surveys Tuts., vol. 18, no. 4, pp. 2847–2886, 4th Quart., 2016.
HetNet,” IEEE Wireless Commun. Mag., vol. 19, no. 2, pp. 4–5, [46] “Mobile-edge computing: Introductory technical white paper,” ETSI,
Apr. 2012. Sophia Antipolis, France, White Paper, Sep. 2014.
[47] V. Jacobson et al., “Networking named content,” in Proc. 5th Int. [73] H. Ahlehagh and S. Dey, “Video caching in radio access network:
Conf. Emerg. Netw. Exp. Technol. (CoNEXT), Rome, Italy, Dec. 2009, Impact on delay and capacity,” in Proc. IEEE Wireless Commun. Netw.
pp. 1–12. Conf. (WCNC), Shanghai, China, Apr. 2012, pp. 2276–2281.
[48] R. Buyya and M. Pathan, Content Delivery Network. Berlin, Germany: [74] N. Golrezaei, K. Shanmugam, A. G. Dimakis, A. F. Molisch, and
Springer, 2008. G. Caire, “FemtoCaching: Wireless video content delivery through
[49] S. Androutsellis-Theotokis and D. Spinellis, “A survey of peer-to- distributed caching helpers,” in Proc. IEEE INFOCOM, Orlando, FL,
peer content distribution technologies,” ACM Comput. Surveys, vol. 36, USA, Mar. 2012, pp. 1107–1115.
no. 4, pp. 335–371, Dec. 2004. [75] L. Han, B. Nath, L. Iftode, and S. Muthukrishnan, “Social butterfly:
[50] A. Passarella, “A survey on content-centric technologies for the current Social caches for distributed social networks,” in Proc. IEEE 3rd Int.
Internet: CDN and P2P solutions,” J. Comput. Commun., vol. 35, no. 1, Conf. Privacy Security Risk Trust IEEE 3rd Int. Conf. Soc. Comput.,
pp. 1–32, Jan. 2011. Boston, MA, USA, Oct. 2011, pp. 81–86.
[51] E. K. Lua, J. Crowcroft, M. Pias, R. Sharma, and S. Lim, “A survey and [76] L. Han, M. Punceva, B. Nath, S. Muthukrishnan, and L. Iftode,
comparison of peer-to-peer overlay network schemes,” IEEE Commun. “SocialCDN: Caching techniques for distributed social networks,” in
Surveys Tuts., vol. 7, no. 2, pp. 72–93, 2nd Quart., 2005. Proc. IEEE Int. Conf. Peer Peer Comput. (P2P), Tarragona, Spain,
[52] B. Ahlgren, C. Dannewitz, C. Imbrenda, D. Kutscher, and B. Ohlman, Sep. 2012, pp. 191–202.
“A survey of information-centric networking,” IEEE Commun. Mag., [77] C. Zhang, Y. Sun, Y. Mo, Y. Zhang, and S. Bu, “Social-aware con-
vol. 50, no. 7, pp. 26–36, Jul. 2012. tent downloading for fog radio access networks supported device-to-
device communications,” in Proc. IEEE Int. Conf. Ubiquitous Wireless
[53] G. Xylomenos et al., “A survey of information-centric networking
Broadband (ICUWB), Nanjing, China, Oct. 2016, pp. 1–4.
research,” IEEE Commun. Surveys Tuts., vol. 16, no. 2, pp. 1024–1049,
[78] O. Semiari, W. Saad, S. Valentin, M. Bennis, and H. V. Poor, “Context-
2nd Quart., 2014.
aware small cell networks: How social metrics improve wireless
[54] K. Djemame, D. Armstrong, J. Guitart, and M. Macias, “A risk assess- resource allocation,” IEEE Trans. Wireless Commun., vol. 14, no. 11,
ment framework for cloud computing,” IEEE Trans. Cloud Comput., pp. 5927–5940, Nov. 2015.
vol. 4, no. 3, pp. 265–278, Jul./Sep. 2016. [79] E. Zeydan et al., “Big data caching for networking: Moving from cloud
[55] A. ur Rehman Khan, M. Othman, S. A. Madani, and S. U. Khan, to edge,” IEEE Commun. Mag., vol. 54, no. 9, pp. 36–42, Sep. 2016.
“A survey of mobile cloud computing application models,” IEEE [80] J. Tang and T. Q. S. Quek, “The role of cloud computing in content-
Commun. Surveys Tuts., vol. 16, no. 1, pp. 393–413, 1st Quart., 2014. centric mobile networking,” IEEE Commun. Mag., vol. 54, no. 8,
[56] M. Zhanikeev, “A cloud visitation platform to facilitate cloud federation pp. 52–59, Aug. 2016.
and fog computing,” Computer, vol. 48, no. 5, pp. 80–83, May 2015. [81] J. Tang, W. P. Tay, and T. Q. S. Quek, “Cross-layer resource allocation
[57] S. Sarkar, S. Chatterjee, and S. Misra, “Assessment of the suitability with elastic service scaling in cloud radio access network,” IEEE Trans.
of fog computing in the context of Internet of Things,” IEEE Trans. Wireless Commun., vol. 14, no. 9, pp. 5068–5081, Sep. 2015.
Cloud Comput., to be published. [82] R. Tandon and O. Simeone, “Harnessing cloud and edge synergies:
[58] R. Gargees et al., “Incident-supporting visual cloud computing uti- Toward an information theory of fog radio access networks,” IEEE
lizing software-defined networking,” IEEE Trans. Circuits Syst. Video Commun. Mag., vol. 54, no. 8, pp. 44–50, Aug. 2016.
Technol., vol. 27, no. 1, pp. 182–197, Jan. 2017. [83] Q. Li, H. Niu, A. Papathanassiou, and G. Wu, “Edge cloud and underlay
[59] N. Iotti, M. Picone, S. Cirani, and G. Ferrari, “Improving quality of networks: Empowering 5G cell-less wireless architecture,” in Proc. Eur.
experience in future wireless access networks through fog computing,” Wireless Conf., Barcelona, Spain, May 2014, pp. 1–6.
IEEE Internet Comput., vol. 21, no. 2, pp. 26–33, Mar./Apr. 2017. [84] S.-H. Park, O. Simeone, and S. S. Shitz, “Joint optimization of cloud
[60] Cisco. (Jan. 2014). Cisco Delivers Vision of Fog Computing to and edge processing for fog radio access networks,” IEEE Trans.
Accelerate Value From Billions of Connected Devices. [Online]. Wireless Commun., vol. 15, no. 11, pp. 7621–7632, Nov. 2016.
Available: http://newsroom.cisco.com/release/1334100/CiscoDelivers- [85] J. A. Khan and Y. Ghamri-Doudane, “SAVING: Socially aware vehic-
Vision-of-Fog-Computing-to-Accelerate-Value-from-Billionsof- ular information-centric networking,” IEEE Commun. Mag., vol. 54,
Connected-Devices-utm-medium-rss no. 8, pp. 100–107, Aug. 2016.
[61] M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies, “The case for [86] Y. Chen, S. Zhang, S. Xu, and G. Y. Li, “Fundamental trade-offs
VM-based cloudlets in mobile computing,” IEEE Pervasive Comput., on green wireless networks,” IEEE Commun. Mag., vol. 49, no. 6,
vol. 8, no. 4, pp. 14–23, Oct./Dec. 2009. pp. 30–37, Jun. 2011.
[62] H. Li, G. Shou, Y. Hu, and Z. Guo, “Mobile edge computing: Progress [87] C. Colman-Meixner, C. Develder, M. Tornatore, and B. Mukherjee,
and challenges,” in Proc. 4th IEEE Int. Conf. Mobile Cloud Comput. “A survey on resiliency techniques in cloud computing infrastruc-
Services Eng., Oxford, U.K., Mar./Apr. 2016, pp. 83–84. tures and applications,” IEEE Commun. Surveys Tuts., vol. 18, no. 3,
[63] X. Chen, L. Jiao, W. Li, and X. Fu, “Efficient multi-user computation pp. 2244–2281, 3rd Quart., 2016.
offloading for mobile-edge cloud computing,” IEEE/ACM Trans. Netw., [88] S. Bu, F. R. Yu, Y. Cai, and X. P. Liu, “When the smart grid meets
vol. 24, no. 5, pp. 2795–2808, Oct. 2016. energy-efficient communications: Green wireless cellular networks
[64] The Mobile Cloud: Market Analysis and Forecasts, ARCchart, London, powered by the smart grid,” IEEE Trans. Wireless Commun., vol. 11,
U.K., 2011. no. 8, pp. 3014–3024, Aug. 2012.
[65] “Cisco visual networking index: Forecast and methodology [89] F. Capozzi, G. Piro, L. A. Grieco, G. Boggia, and P. Camarda,
2012–2017,” CISCO, San Jose, CA, USA, White Paper, May 2013. “Downlink packet scheduling in LTE cellular networks: Key design
issues and a survey,” IEEE Commun. Surveys Tuts., vol. 15, no. 2,
[66] G. Tyson et al., “A trace-driven analysis of caching in content-centric
pp. 678–700, 2nd Quart., 2013.
networks,” in Proc. IEEE 21st Int. Conf. Comput. Commun. Netw.,
[90] “Technical specification group services and system aspects; pol-
Munich, Germany, Aug. 2012, pp. 1–7.
icy and charging control architecture,” 3rd Gener. Partnership
[67] Juniper Research. (Jul. 2011). Mobile Cloud: Smart Device Strategies Project, Sophia Antipolis, France, Tech. Rep. TS 23.203 V12.2.0,
for Enterprise & Consumer Markets 2011–2016. [Online]. Available: Sep. 2013. [Online]. Available: http://www.3gpp.org/ftp/specs/html-
http://juniperresearch.com/ INFO/23203.htm
[68] M. Sheng et al., “Enhancement for content delivery with proximity [91] Y. Li, H. Xie, Y. Wen, C.-Y. Chow, and Z.-L. Zhang, “How
communications in caching enabled wireless networks: Architecture much to coordinate? Optimizing in-network caching in content-
and challenges,” IEEE Commun. Mag., vol. 54, no. 8, pp. 70–76, centric networks,” IEEE Trans. Netw. Service Manag., vol. 12, no. 3,
Aug. 2016. pp. 420–434, Sep. 2015.
[69] J. Erman et al., “To cache or not to cache: The 3G case,” IEEE Internet [92] D. Huang, P. Wang, and D. Niyato, “A dynamic offloading algorithm
Comput., vol. 15, no. 2, pp. 27–34, Mar./Apr. 2011. for mobile computing,” IEEE Trans. Wireless Commun., vol. 11, no. 6,
[70] S. Woo et al., “Comparison of caching strategies in modern cellular pp. 1991–1995, Jun. 2012.
backhaul networks,” in Proc. 11th ACM Annu. Int. Conf. Mobile Syst. [93] O. Muñoz, A. Pascual-Iserte, and J. Vidal, “Optimization of radio and
Appl. Services (MobiSys), Taipei, Taiwan, Jun. 2013, pp. 319–332. computational resources for energy efficiency in latency-constrained
[71] X. Wang, M. Chen, T. Taleb, A. Ksentini, and V. Leung, “Cache in application offloading,” IEEE Trans. Veh. Technol., vol. 64, no. 10,
the air: Exploiting content caching and delivery techniques for 5G pp. 4738–4755, Oct. 2015.
systems,” IEEE Commun. Mag., vol. 52, no. 2, pp. 131–139, Feb. 2014. [94] W. Zhang et al., “Energy-optimal mobile cloud computing under
[72] “Network functions virtualisation (NFV): Use cases,” ETSI, stochastic wireless channel,” IEEE Trans. Wireless Commun., vol. 12,
Sophia Antipolis, France, Tech. Rep. 2013001, Mar. 2013. no. 9, pp. 4569–4581, Sep. 2013.
[95] S. Sardellitti, G. Scutari, and S. Barbarossa, “Joint optimization of radio [118] C. Li and H. Dai, “Efficient in-network computing with noisy
and computational resources for multicell mobile-edge computing,” wireless channels,” IEEE Trans. Mobile Comput., vol. 12, no. 11,
IEEE Trans. Signal Inf. Process. Over Netw., vol. 1, no. 2, pp. 89–103, pp. 2167–2177, Nov. 2013.
Jun. 2015. [119] C. You, K. Huang, and H. Chae, “Energy efficient mobile cloud
[96] J. Kwak, Y. Kim, J. Lee, and S. Chong, “DREAM: Dynamic computing powered by wireless energy transfer,” IEEE J. Sel. Areas
resource and task allocation for energy minimization in mobile cloud Commun., vol. 34, no. 5, pp. 1757–1771, May 2016.
systems,” IEEE J. Sel. Areas Commun., vol. 33, no. 12, pp. 2510–2523, [120] M. Kangas, S. Glisic, Y. Fang, and P. Li, “Resource harvesting in
Dec. 2015. cognitive wireless computing networks with mobile clouds and virtu-
[97] X. Chen, “Decentralized computation offloading game for mobile alized distributed data centers: Performance limits,” IEEE Trans. Cogn.
cloud computing,” IEEE Trans. Parallel Distrib. Syst., vol. 26, no. 4, Commun. Netw., vol. 1, no. 3, pp. 318–334, Sep. 2015.
pp. 974–983, Apr. 2015. [121] X. Liu, Y. Li, and H.-H. Chen, “Wireless resource scheduling based
[98] Y. Cai, F. R. Yu, and S. Bu, “Dynamic operations of cloud radio access on backoff for multiuser multiservice mobile cloud computing,” IEEE
networks (C-RAN) for mobile cloud computing systems,” IEEE Trans. Trans. Veh. Technol., vol. 65, no. 11, pp. 9247–9259, Nov. 2016.
Veh. Technol., vol. 65, no. 3, pp. 1536–1548, Mar. 2016. [122] R. C. Chiang, S. Rajasekaran, N. Zhang, and H. H. Huang, “Swiper:
[99] Y. Mao, J. Zhang, and K. B. Letaief, “Dynamic computation offloading Exploiting virtual machine vulnerability in third-party clouds with com-
for mobile-edge computing with energy harvesting devices,” IEEE J. petition for I/O resources,” IEEE Trans. Parallel Distrib. Syst., vol. 26,
Sel. Areas Commun., vol. 34, no. 12, pp. 3590–3605, Dec. 2016. no. 6, pp. 1732–1742, Jun. 2015.
[100] N. Kumar, S. Zeadally, and J. J. P. C. Rodrigues, “Vehicular delay- [123] R. Deng, R. Lu, C. Lai, T. H. Luan, and H. Liang, “Optimal workload
tolerant networks for smart grid data management using mobile allocation in fog-cloud computing toward balanced delay and power
edge computing,” IEEE Commun. Mag., vol. 54, no. 10, pp. 60–66, consumption,” IEEE Internet Things J., vol. 3, no. 6, pp. 1171–1181,
Oct. 2016. Dec. 2016.
[101] K. Poularakis and L. Tassiulas, “Code, cache and deliver on the move: [124] L. Pu, X. Chen, J. Xu, and X. Fu, “D2D fogging: An energy-efficient
A novel caching paradigm in hyper-dense small-cell networks,” IEEE and incentive-aware task offloading framework via network-assisted
Trans. Mobile Comput., vol. 16, no. 3, pp. 675–687, Mar. 2017. D2D collaboration,” IEEE J. Sel. Areas Commun., vol. 34, no. 12,
[102] A. Khreishah, J. Chakareski, and A. Gharaibeh, “Joint caching, routing, pp. 3887–3901, Dec. 2016.
and channel assignment for collaborative small-cell cellular networks,” [125] X. Hou et al., “Vehicular fog computing: A viewpoint of vehicles
IEEE J. Sel. Areas Commun., vol. 34, no. 8, pp. 2275–2284, Aug. 2016. as the infrastructures,” IEEE Trans. Veh. Technol., vol. 65, no. 6,
[103] D. Das and A. A. Abouzeid, “Co-operative caching in dynamic shared pp. 3860–3873, Jun. 2016.
spectrum networks,” IEEE Trans. Wireless Commun., vol. 15, no. 7, [126] F. Jalali, K. Hinton, R. Ayre, T. Alpcan, and R. S. Tucker, “Fog com-
pp. 5060–5075, Jul. 2016. puting may help to save energy in cloud computing,” IEEE J. Sel. Areas
[104] D. Liu and C. Yang, “Energy efficiency of downlink networks with Commun., vol. 34, no. 5, pp. 1728–1739, May 2016.
caching at base stations,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, [127] D. Zeng, L. Gu, S. Guo, Z. Cheng, and S. Yu, “Joint optimization
pp. 907–922, Apr. 2016. of task scheduling and image placement in fog computing supported
[105] S. Wang, J. Bi, J. Wu, and A. V. Vasilakos, “CPHR: In-network caching software-defined embedded system,” IEEE Trans. Comput., vol. 65,
for information-centric networking with partitioning and hash-routing,” no. 12, pp. 3702–3712, Dec. 2016.
IEEE/ACM Trans. Netw., vol. 24, no. 5, pp. 2742–2755, Oct. 2016. [128] S. Cirani, G. Ferrari, N. Iotti, and M. Picone, “The IoT hub:
[106] X. Yuan et al., “Enabling secure and efficient video delivery through A fog node for seamless management of heterogeneous connected
encrypted in-network caching,” IEEE J. Sel. Areas Commun., vol. 34, smart objects,” in Proc. 12th Annu. IEEE Int. Conf. Sens. Commun.
no. 8, pp. 2077–2090, Aug. 2016. Netw. Workshops (SECON Workshops), Seattle, WA, USA, Jun. 2015,
[107] P. Si, H. Yue, Y. Zhang, and Y. Fang, “Spectrum management for proac- pp. 1–6.
tive video caching in information-centric cognitive radio networks,” [129] G. Lee, W. Saad, and M. Bennis, “An online secretary framework for
IEEE J. Sel. Areas Commun., vol. 34, no. 8, pp. 2247–2259, Aug. 2016. fog network formation with minimal latency,” in Proc. IEEE Int. Conf.
[108] D. Malak, M. Al-Shalash, and J. Andrews, “Optimizing content caching Commun. (ICC), Paris, France, May 2017, pp. 1–6.
to maximize the density of successful receptions in device-to-device [130] Y. Wang, M. Sheng, X. Wang, L. Wang, and J. Li, “Mobile-edge com-
networking,” IEEE Trans. Commun., vol. 64, no. 10, pp. 4365–4380, puting: Partial computation offloading using dynamic voltage scaling,”
Oct. 2016. IEEE Trans. Commun., vol. 64, no. 10, pp. 4268–4282, Oct. 2016.
[109] L. Zhang, M. Xiao, G. Wu, and S. Li, “Efficient scheduling and power [131] C. Wang, F. R. Yu, C. Liang, Q. Chen, and L. Tang, “Joint computation
allocation for D2D-assisted wireless caching networks,” IEEE Trans. offloading and interference management in wireless cellular networks
Commun., vol. 64, no. 6, pp. 2438–2452, Jun. 2016. with mobile edge computing,” IEEE Trans. Veh. Technol., vol. 66, no. 8,
[110] M. Gregori, J. Gómez-Vilardeb, J. Matamoros, and D. Gündüz, pp. 7432–7445, Aug. 2017.
“Wireless content caching for small cell and D2D networks,” IEEE [132] C. Wang, F. R. Yu, Q. Chen, and L. Tang, “Joint computation and
J. Sel. Areas Commun., vol. 34, no. 5, pp. 1222–1234, May 2016. radio resource management for cellular networks with mobile edge
[111] A. Liu and V. K. N. Lau, “Exploiting base station caching in MIMO computing,” in Proc. IEEE ICC, Paris, France, May 2017, pp. 1–6.
cellular networks: Opportunistic cooperation for video streaming,” [133] N. Master, A. Dua, D. Tsamis, J. P. Singh, and N. Bambos, “Adaptive
IEEE Trans. Signal Process., vol. 63, no. 1, pp. 57–69, Jan. 2015. prefetching in wireless computing,” IEEE Trans. Wireless Commun.,
[112] K. Poularakis, G. Iosifidis, I. Pefkianakis, L. Tassiulas, and M. May, vol. 15, no. 5, pp. 3296–3310, May 2016.
“Mobile data offloading through caching in residential 802.11 wire- [134] Y. Jin, Y. Wen, and C. Westphal, “Optimal transcoding and caching
less networks,” IEEE Trans. Netw. Service Manag., vol. 13, no. 1, for adaptive streaming in media cloud: An analytical approach,” IEEE
pp. 71–84, Mar. 2016. Trans. Circuits Syst. Video Technol., vol. 25, no. 12, pp. 1914–1925,
[113] S. H. Chae and W. Choi, “Caching placement in stochastic wireless Dec. 2015.
caching helper networks: Channel selection diversity via caching,” [135] G. Gao, W. Zhang, Y. Wen, Z. Wang, and W. Zhu, “Towards cost-
IEEE Trans. Wireless Commun., vol. 15, no. 10, pp. 6626–6637, efficient video transcoding in media cloud: Insights learned from
Oct. 2016. user viewing patterns,” IEEE Trans. Multimedia, vol. 17, no. 8,
[114] C. Zhu et al., “A novel sensory data processing framework to integrate pp. 1286–1296, Aug. 2015.
sensor networks with mobile cloud,” IEEE Syst. J., vol. 10, no. 3, [136] A. Khanafer, M. Kodialam, and K. P. N. Puttaswamy, “To rent or to buy
pp. 1125–1136, Sep. 2016. in the presence of statistical information: The constrained ski-rental
[115] P. Zhang, Z. Yan, and H. Sun, “A novel architecture based on cloud problem,” IEEE/ACM Trans. Netw., vol. 23, no. 4, pp. 1067–1077,
computing for wireless sensor network,” in Proc. 2nd Int. Conf. Aug. 2015.
Comput. Sci. Electron. Eng., 2013, pp. 472–475. [137] J. Liu, Y. Zhou, and D. Zhang, “Transim: A simulation framework for
[116] C. Zhu, V. C. M. Leung, L. T. Yang, and L. Shu, “Collaborative cache-enabled transparent computing systems,” IEEE Trans. Comput.,
location-based sleep scheduling for wireless sensor networks integrat- vol. 65, no. 10, pp. 3171–3183, Oct. 2016.
edwith mobile cloud computing,” IEEE Trans. Comput., vol. 64, no. 7, [138] E. Baştuğ et al., “Big data meets telcos: A proactive caching perspec-
pp. 1844–1856, Jul. 2015. tive,” J. Commun. Netw., vol. 17, no. 6, pp. 549–557, Dec. 2015.
[117] W. Zhang, Y. Wen, and D. O. Wu, “Collaborative task execution in [139] X. Dai, X. Wang, and N. Liu, “Optimal scheduling of data-intensive
mobile cloud computing under a stochastic wireless channel,” IEEE applications in cloud-based video distribution services,” IEEE Trans.
Trans. Wireless Commun., vol. 14, no. 1, pp. 81–93, Jan. 2015. Circuits Syst. Video Technol., vol. 27, no. 1, pp. 73–83, Jan. 2017.
[140] J. Mukherjee, D. Krishnamurthy, and J. Rolia, “Resource contention [164] Y. He et al., “Big data analytics in mobile cellular networks,” IEEE
detection in virtualized environments,” IEEE Trans. Netw. Service Access, vol. 4, pp. 1985–1996, 2016.
Manag., vol. 12, no. 2, pp. 217–231, Jun. 2015. [165] O. F. Celebi et al., “On use of big data for enhancing network coverage
[141] M. K. Kiskani and H. R. Sadjadpour, “Throughput analysis of decen- analysis,” in Proc. ICT, Casablanca, Morocco, May 2013, pp. 1–5.
tralized coded content caching in cellular networks,” IEEE Trans. [166] J. Liu, N. Chang, S. Zhang, and Z. Lei, “Recognizing and character-
Wireless Commun., vol. 16, no. 1, pp. 663–672, Jan. 2017. izing dynamics of cellular devices in cellular data network through
[142] C. Wang, C. Liang, F. R. Yu, Q. Chen, and L. Tang, “Computation massive data analysis,” Int. J. Commun. Syst., vol. 28, no. 12,
offloading and resource allocation in wireless cellular networks with pp. 1884–1897, 2015.
mobile edge computing,” IEEE Trans. Wireless Commun., vol. 16, [167] V. Mnih et al., “Human-level control through deep reinforcement
no. 8, pp. 4924–4938, Aug. 2017. learning,” Nature, vol. 518, no. 7540, pp. 529–533, 2015.
[143] C. Wang, C. Liang, F. R. Yu, Q. Chen, and L. Tang, “Joint compu- [168] D. Silver et al., “Mastering the game of go with deep neural networks
tation offloading, resource allocation and content caching in cellular and tree search,” Nature, vol. 529, no. 7587, pp. 484–489, 2016.
networks with mobile edge computing,” in Proc. IEEE ICC, Paris, [169] Y. He et al., “Deep reinforcement learning-based optimization for
France, May 2017, pp. 1–6. cache-enabled opportunistic interference alignment wireless networks,”
[144] M. Chiang and T. Zhang, “Fog and IoT: An overview of research IEEE Trans. Veh. Technol., to be published.
opportunities,” IEEE Internet Things J., vol. 3, no. 6, pp. 854–864, [170] T. Y. He, N. Zhao, and H. Yin, “Integrated networking, caching
Dec. 2016. and computing for connected vehicles: A deep reinforcement learning
[145] J.-Q. Li et al., “Industrial Internet: A survey on the enabling tech- approach,” IEEE Trans. Veh. Technol., to be published.
nologies, applications, and challenges,” IEEE Commun. Surveys Tuts.,
vol. 19, no. 3, pp. 1504–1526, 3rd Quart., 2017.
[146] M. Tariq, M. S. A. Latiff, M. Ayaz, Y. Coulibaly, and A. Wahid,
“Pressure sensor based reliable (PSBR) routing protocol for underwa-
Chenmeng Wang received the M.S. degree in
ter acoustic sensor networks,” Ad Hoc Sensor Wireless Netw., vol. 32,
information and telecommunication engineering
nos. 3–4, pp. 175–196, 2016.
from the Chongqing University of Posts and
[147] C. Wang and Y. Zhang, “Time-window and voronoi-partition based
Telecommunications, Chongqing, China, in 2014,
aggregation scheduling in multi-sink wireless sensor networks,” Ad Hoc
where he is currently pursuing the Ph.D. degree
Sensor Wireless Netw., vol. 32, nos. 3–4, pp. 221–238, 2016.
with the School of Communication and Information
[148] R. Kelly. (2017). Internet of Things Data to Top 1.6 Zettabytes
Engineering. From 2015 to 2017, he was a visit-
by 2022. [Online]. Available: https://campustechnology.com/articles/
ing student with Carleton University, Ottawa, ON,
2015/04/15/internet-of-thingsdata-to-top-1-6-zettabytes-by-2020.aspx
Canada. His current research interests include small
[149] L. Mearian. (2017) Self-Driving Cars Could Create 1GB of Data
cell networks, mobile edge computing systems,
a Second. [Online]. Available: http://www.computerworld.com/
resource allocation, and applications of convex opti-
article/2484219/emerging-technology/self-driving-cars-could-create-
mization in mobile networks.
1gb-of-data-a-second.html
[150] N. Cochrane. (2017). U.S. Smart Grid to Generate 1000 Petabytes
of Data a Year. [Online]. Available: http://www.itnews.com.au/
news/us-smart-grid-to-generate-1000-petabytes-of-data-a-year-
170290ixzz458VaITi6 Ying He (S’16) received the B.S. degree in com-
[151] D. Gesbert et al., “Multi-cell MIMO cooperative networks: A new munication and information systems from Dalian
look at interference,” IEEE J. Sel. Areas Commun., vol. 28, no. 9, Ocean University, Dalian, China, in 2011 and the
pp. 1380–1408, Dec. 2010. M.S. degree in communication and information
[152] R. Tourani, S. Misra, and T. Mick, “IC-MCN: An architecture for an systems from the Dalian University of Technology,
information-centric mobile converged network,” IEEE Commun. Mag., Dalian, in 2015. She is currently pursuing the Ph.D.
vol. 54, no. 9, pp. 43–49, Sep. 2016. degree with the Dalian University of Technology and
[153] A. S. Gomes et al., “Edge caching with mobility prediction in virtu- Carleton University. Her current research interests
alized LTE mobile networks,” Future Gener. Comput. Syst., vol. 70, include big data, security, wireless networks, and
pp. 148–162, May 2017. machine learning.
[154] Y. Rao, H. Zhou, D. Gao, H. Luo, and Y. Liu, “Proactive caching for
enhancing user-side mobility support in named data networking,” in
Proc. Int. Conf. Innov. Mobile Internet Services Ubiquitous Comput.,
Taichung, Taiwan, Jul. 2013, pp. 37–42. F. Richard Yu (S’00–M’04–SM’08) received the
[155] O. Semiari, W. Saad, M. Bennis, and B. Maham. Caching Meets Ph.D. degree in electrical engineering from the
Millimeter Wave Communications for Enhanced Mobility Management University of British Columbia in 2003. From 2002
in 5G Networks. [Online]. Available: https://arxiv.org/pdf/1701.05125 to 2006, he was with Ericsson, Lund, Sweden,
[156] H. Liu, Z. Chen, X. Tian, X. Wang, and M. Tao, “On content-centric and a start-up in California, USA. He joined
wireless delivery networks,” IEEE Wireless Commun., vol. 21, no. 6, Carleton University in 2007, where he is currently
pp. 118–125, Dec. 2014. a Professor. His research interests include cross-
[157] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung, “Network infor- layer/cross-system design, connected vehicles, secu-
mation flow,” IEEE Trans. Inf. Theory, vol. 46, no. 4, pp. 1204–1216, rity, and green ICT. He was a recipient of the
Jul. 2000. IEEE Outstanding Service Award in 2016, the IEEE
[158] X.-H. You et al., “Cooperative distributed antenna systems for mobile Outstanding Leadership Award in 2013, the Carleton
communications [coordinated and distributed MIMO],” IEEE Wireless Research Achievement Award in 2012, the Ontario Early Researcher Award
Commun., vol. 17, no. 3, pp. 35–43, Jun. 2010. (formerly, Premiers Research Excellence Award) in 2011, the Excellent
[159] H. Hu, H.-H. Chen, P. Mueller, R. Q. Hu, and Y. Rui, “Software defined Contribution Award at IEEE/IFIP TrustCom 2010, the Leadership Opportunity
wireless networks (SDWN): Part 1 [guest editorial],” IEEE Commun. Fund Award from Canada Foundation of Innovation in 2009, and the Best
Mag., vol. 53, no. 11, pp. 108–109, Nov. 2015. Paper Awards at IEEE VTC 2017 Spring, ICC 2014, Globecom 2012,
[160] K. Wang, H. Li, F. R. Yu, and W. Wei, “Virtual resource allocation in IEEE/IFIP TrustCom 2009 and Int’l Conference on Networking 2005.
software-defined information-centric cellular networks with device-to- He serves on the editorial boards of several journals, including the Co-
device communications and imperfect CSI,” IEEE Trans. Veh. Technol., Editor-in-Chief for Ad Hoc & Sensor Wireless Networks, a Lead Series Editor
vol. 65, no. 12, pp. 10011–10021, Dec. 2016. for the IEEE T RANSACTIONS ON V EHICULAR T ECHNOLOGY, the IEEE
[161] Y. Li and M. Chen, “Software-defined network function virtualization: T RANSACTIONS ON G REEN C OMMUNICATIONS AND N ETWORKING, and
A survey,” IEEE Access, vol. 3, pp. 2542–2553, 2015. the IEEE C OMMUNICATIONS S URVEYS & T UTORIALS. He has served as
[162] J. Matias, J. Garay, N. Toledo, J. Unzilla, and E. Jacob, “Toward an the Technical Program Committee Co-Chair of numerous conferences. He
SDN-enabled NFV architecture,” IEEE Commun. Mag., vol. 53, no. 4, is a registered Professional Engineer in the province of Ontario, Canada,
pp. 187–193, Apr. 2015. and a fellow of the Institution of Engineering and Technology. He is a
[163] H. Wen, P. K. Tiwary, and T. Le-Ngoc, Wireless Virtualization. Cham, Distinguished Lecturer and a member of Board of Governors of the IEEE
Switzerland: Springer, 2013. Vehicular Technology Society.
Qianbin Chen received the Ph.D. degree in commu- Lun Tang received the Ph.D. degree in commu-
nication and information system from the University nication and information system from Chongqing
of Electronic Science and Technology of China, University, Chongqing, China. He is currently a
Chengdu, China, in 2002. He is currently a Professor Professor with the School of Communication and
with the School of Communication and Information Information Engineering, Chongqing University of
Engineering, Chongqing University of Posts and Posts and Telecommunications. His current research
Telecommunications, where he is the Director interests include 5G cellular network, interference
of the Chongqing Key Laboratory of Mobile management, and small cell networks.
Communication Technology. He has published over
100 papers in journals and peer-reviewed conference
proceedings, and has co-authored 7 books. He holds
47 granted national patents.

Integration of Networking, Caching, and Computing in Wireless Systems: A Survey, Some Research Issues, and Challenges

Uploaded by

Copyright:

Available Formats

Integration of Networking, Caching, and Computing in Wireless Systems: A Survey, Some Research Issues, and Challenges

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

Integration of Networking, Caching, and Computing in Wireless Systems: A Survey, Some Research Issues, and Challenges

Uploaded by

Copyright:

Available Formats

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

1, FIRST QUARTER 2018 7

Integration of Networking, Caching, and Computing

of the incorporation of fog computing into wireless

Fig. 4. Signaling interaction of the networking-caching framework [68].

service providers [72]. In this approach, several services share

in small cell networks with D2D communication is proposed.

Cleansing, parsing and formatting data: In order to per-

Fig. 9. Caching-computing architecture [79].

in SDN, which facilitates to guarantee flexibility of the

vehicles to relay and retrieve cached content, in order to

A. General Metrics B. Networking-Related Metrics

You might also like