Cloud-Based Augmentation for Mobile Devices: Motivation, Taxonomies, and Open Challenges

IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 1
Cloud-Based Augmentation for Mobile Devices:
Motivation, Taxonomies, and Open Challenges
Saeid Abolfazli, Member, IEEE, Zohreh Sanaei, Member, IEEE, Ejaz Ahmed, Member, IEEE, Abdullah
Gani, Senior Member, IEEE, Rajkumar Buyya, Senior Member, IEEE
Abstract—Recently, Cloud-based Mobile Augmentation (CMA)
approaches have gained remarkable ground from academia and
industry. CMA is the state-of-the-art mobile augmentation model
that employs resource-rich clouds to increase, enhance, and
optimize computing capabilities of mobile devices aiming at
execution of resource-intensive mobile applications. Augmented
mobile devices envision to perform extensive computations and
to store big data beyond their intrinsic capabilities with least
footprint and vulnerability. Researchers utilize varied cloud-
based computing resources (e.g., distant clouds and nearby
mobile nodes) to meet various computing requirements of mobile
users. However, employing cloud-based computing resources is
not a straightforward panacea. Comprehending critical factors
(e.g., current state of mobile client and remote resources) that
impact on augmentation process and optimum selection of cloud-
based resource types are some challenges that hinder CMA
adaptability. This paper comprehensively surveys the mobile aug-
mentation domain and presents taxonomy of CMA approaches.
The objectives of this study is to highlight the effects of remote
resources on the quality and reliability of augmentation processes
and discuss the challenges and opportunities of employing varied
cloud-based resources in augmenting mobile devices. We present
augmentation de?nition, motivation, and taxonomy of augmen-
tation types, including traditional and cloud-based. We critically
analyze the state-of-the-art CMA approaches and classify them
into four groups of distant ?xed, proximate ?xed, proximate
mobile, and hybrid to present a taxonomy. Vital decision making
and performance limitation factors that in?uence on the adoption
of CMA approaches are introduced and an exemplary decision
making ?owchart for future CMA approaches are presented. Im-
pacts of CMA approaches on mobile computing is discussed and
open challenges are presented as the future research directions.
Index Terms—Cloud-based Mobile Augmentation, Mobile
Cloud Computing, Cloud Computing, Resource-intensive Mobile
Application, Computation Of?oading, Resource Outsourcing.
I. INTRODUCTION
S
INCE a decade ago, the visions of ‘information under
?ngertip’ [1] and ‘unrestricted mobile computing’ [2]
aroused the need to enhance computing power of mobile
devices to meet the insatiable computing demands of mobile
users [3]. In the late 90s, the concept of load sharing and
Manuscript received Dec 18, 2012; revised March 05, 2013 and 06
May, 2013;This work is funded by the Malaysian Ministry of Higher
Education under the University of Malaya High Impact Research Grant -
UM.C/HIR/MOHE/FCSIT/03. Ejaz Ahmed’s research work is supported by
the Bright Spark Unit, University of Malaya, Malaysia.
Saeid Abolfazli(corresponding author), Zohreh Sanaei, Ejaz Ahmed, and
Abdullah Gani are with the Department of Computer System & Technology,
The University of Malaya, Kuala Lumpur, Malaysia (e-mail: {abolfazli,sanaei,
ejazahmed}@ieee.org; [email protected])
RajKumar Buyya is with the Department of Computing and Information
Systems, The University of Melbourne, 111, Barry Street, Carlton, Melbourne,
VIC 3053, Australia, Email: [email protected]
remote execution aimed to augment computing capabilities of
mobile devices by shifting the resource-intensive mobile codes
to surrogates (powerful computing device(s) in vicinity) [4]–
[6]. Although remote execution efforts [7]–[18] have yielded
many impressive achievements, several challenges such as
reliability, security, and elasticity of surrogates hinder the
remote execution adaptability [19]. For instance, the resource
sharing and computing services of surrogates can be termi-
nated without prior notice and their content can be accessed
and altered by the surrogate machine or other users in the
absence of a Service Level Agreement (SLA). SLA is a formal
contract employed and negotiated in advance between service
provider and consumer to enforce certain level of quality
against a fee.
Few years later, emergence of cloud resources created an
opportunity to mitigate the shortcomings of utilizing surro-
gates in augmenting mobile devices. Cloud is a type of dis-
tributed system comprised of a cluster of powerful computers
accessible as uni?ed computing resource(s) based on an SLA
[20]. Cloud computing as “a model for enabling ubiquitous,
convenient, on-demand network access to a shared pool of con-
?gurable computing resources (e.g., networks, servers, storage,
applications, and services) that can be rapidly provisioned
and released with minimal management effort or service or
service provider interaction” [21] stimulates researchers to
adopt the cutting edge technology in mobile device augmenta-
tion: Cloud-based Mobile Augmentation (CMA). Cloud-based
Mobile Augmentation (CMA) is the-state-of-the-art mobile
augmentation model that leverages cloud computing technolo-
gies and principles to increase, enhance, and optimize com-
puting capabilities of mobile devices by executing resource-
intensive mobile application components in the resource-rich
cloud-based resources. Cloud-based resources include varied
types of mobile/immobile computing devices that follow cloud
computing principles [22], [23] to perform computations on
behalf of the resource-constraint mobile devices. Figure 1
depicts major building blocks of a typical CMA system. It
is notable that these building blocks are optional superset, and
speci?c CMA system may not have all these building blocks.
CMA efforts [24]–[27], [27]–[49] exploit various cloud-
based computing resources, especially distant clouds and prox-
imate mobile nodes to augment mobile devices. Distant clouds
are giant clouds such as Amazon EC2
1
located inside the
vendor premise —far away from the mobile clients—offering
in?nite, elastic computing resources with extreme computing
1http://aws.amazon.com/ec2/
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 2
Fig. 1. Major Building Blocks of an Exemplary CMA System.
power and high WAN (Wide Area Network) latency. Proximate
mobile nodes are building a cluster of mobile computing de-
vices scattered near the mobile clients offer limited computing
power with lower WAN latency than distant clouds.
Although heterogeneity among cloud-based resources in-
creases service ?exibility and enhances users’ computing expe-
rience, determining the most appropriate computing resources
among available options and performing upfront analysis of
in?uential factors (e.g., user preferences and available native
mobile resources) are critical in the adaptability of CMA
approaches. Thus, ‘resource scheduler’ and ‘analyzer and
optimizer’ components depicted in Figure 1 are needed to
analyze and allocate appropriate resources to each task in
a typical CMA system. Moreover, several questions need to
be addressed before the CMA concept can be successfully
employed in the real scenarios. For instance, can CMA aug-
ment computing capabilities of mobile devices and save local
resources to enhance user experience? Is CMA always feasible
and bene?cial? Which type of resources is appropriate for a
certain task? Answering these questions requires ‘monitoring
and pro?ler’, ‘QoS management’, ‘context management’, and
‘decision making engine’ components to perform in each
CMA system (see Figure 1). Therefore, an augmentation deci-
sion engine similar to those used in [25], [33], [49] and exem-
plary decision making ?ow presented in this paper (discussed
in part VI-C) to determine the mobile augmentation feasibility
is needed to amend the CMA performance and reliability.
During augmentation process, the local and native application
state stack needs synchronization to ensure integrity between
native and remote data. Upon successful outsourcing, remote
results need to be returned and integrated to the source mobile
device. Thus, the ‘Synchronizer’ component needs to perform
in typical CMA approaches (see Figure 1).
Although CMA approaches can empower mobile process-
ing and storage capabilities, several disadvantages such as
application development complexity and unauthorized access
to remote data demand a systematized plenary solution.
Performance of the CMA systems is highly in?uenced by
various challenges and issues of wireless networking and
cloud computing technologies. CMA researchers require a
high performance, elastic, robust, reliable, and foreseeable
communication throughput between mobile nodes and cloud
servers which is not yet realized despite of remarkable efforts
and achievements of communication and networking societies.
Current shortcomings and de?ciencies of wireless communi-
cation and networking, especially seamless connectivity and
mobility, high performance communication throughput pro-
visioning, and wireless data interception discourage system
analysts, engineers, developers, and entrepreneurs from de-
ploying CMA-enabled mobile applications due to the high risk
of system malfunction and user experience degradation.
Moreover, CMA systems require accurate estimation mech-
anisms to predict the overall time and energy consumption
of communication and computation tasks while exploiting
clouds. Such estimation is a challenging task considering
huge infrastructures’ performance diversity [50] and policy
heterogeneity [51] of cloud services in intermittent wireless
environment. Despite of blooming efforts endeavoring to ana-
lyze and comprehend the cloud computing model and behavior
[52]–[55], CMA solutions are still unable to accurately fore-
see required time and energy of exploiting cloud resources
to execute intensive applications. Additionally, sundry cloud
challenges, especially live VM migration, infrastructure and
platform heterogeneity, ef?cient allocation of clouds to tasks,
QoS management, security, privacy, and trust in cloud increase
system complexity and decrease successful CMA systems
adoption.
Among limited studies of the domain, [19] and [56] survey
remote execution and application of?oading algorithms with
focus on how task of?oading is performed in various efforts.
Fernando et al. [57] and Dinh et al. [58] sought to explain the
convergence of mobile and cloud computing, and distinguish
it from the earlier domains such as cloud and grid computing
[59]. The authors describe issues, particularly mobile applica-
tion of?oading, privacy and security, context awareness, and
data management. Sanaei, Abolfazli, Gani, and Buyya [51]
present a comprehensive survey on MCC with major focus
on heterogeneity. The authors describe the challenges and
opportunities imposed by heterogeneity and identify hardware,
platform, feature, API, and network as the roots of MCC
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 3
TABLE I
LIST OF ACRONYMS AND CORRESPONDING FULL FORMS.
Acronym Full form
2D 2 Dimensional
2G 2nd Generation
3D 3 Dimensional
3G 3rd Generation
API Application Programming Interface
App Application (mobile application)
ARM Advanced RISC Machines
CMA Cloud-based Mobile Augmentation
CPU Central Processing Unit
DSL Domain Speci?c Language
DVMS Dynamic VM Synthesis
FTP File Transfer Protocol
GPU Graphics Processing Unit
GUI Graphical User Interface
I/O Input/Output
IaaS Infrastructure as a Service
IP Internet Protocol
IP TV Internet Protocol Television
iSCSI Internet Small Computer System Interface
MCC Mobile Cloud Computing
MNO Mobile Network Operator
OS Operating System
P2P Peer-to-Peer
PC Personal Computer
QoS Quality of Service
R&D Research and Development
RAM Random Access Memory
RISC Reduced Instruction Set Computing
RPC Remote Procedure Call
SAL Service Abstraction Layer
SLA Service Level Agreement
TCP Transmission Control Protocol
UDDI Universal Description Discovery and Integration
UI User Interface
VM Virtual Machine
WAN Wide Area Network
Wi-Fi Wireless Fidelity
heterogeneity. They explain major heterogeneity handling ap-
proaches, particularly virtualization, service oriented archi-
tecture, and semantic technology. However, the computing
performance, distance, elasticity, availability, reliability, and
multi-tenancy of remote resources are marginally discussed
in these studies that necessitate further research to explain
the impact of remote resources on augmentation process and
highlight paradigm shift from the unreliable surrogates to
reliable clouds.
In this paper, we survey the state-of-the-art mobile augmen-
tation efforts that employ cloud computing infrastructures to
enhance computing capabilities of resource-constraint mobile
devices, especially smartphones. To the best of our knowledge,
this is the ?rst effort that studies the impacts of cloud-based
computing resources on mobile augmentation process. We dif-
ferentiate augmentation from similar concepts of load sharing
and remote execution, and present augmentation motivation.
We review efforts that endeavor to mitigate the mobile devices’
shortcomings and classify them as hardware and software to
devise a taxonomy. The impacts of CMA in mobile comput-
ing are presented. The characteristics of cloud-based remote
resources and their role in CMA effectiveness are studied and
classi?ed into four groups, namely distant immobile clouds,
proximate immobile computing entities, proximate mobile
computing entities, and hybrid based on their mobility and
physical location traits. Furthermore, the state-of-the-art CMA
models are reviewed and taxonomized into four classes of
distant ?xed, proximate ?xed, proximate mobile, and hybrid
according to our cloud-based resource classi?cation. Factors
impact on the CMA adaptability are identi?ed and described as
augmentation environment, user preferences and requirements,
mobile devices, cloud servers, and contents. Five major metrics
that limit the performance of CMA approaches are studied. A
sample ?owchart of decision making engines for imminent
CMA solutions is presented and several open challenges are
discussed as the future research directions. Such survey is
bene?cial to the communication and networking communities,
because comprehending CMA process and current deploy-
ment challenges are bene?cial in modifying the fundamental
networking infrastructures to optimize the CMA process. In
this paper, we use the terms mobile devices and smartphones
interchangeably with similar notion. Table I shows the list of
acronyms used in the paper.
The remainder of this paper is organized as follows. Section
II introduces mobile computation augmentation, presents its
motivation and describes the taxonomy of mobile augmenta-
tion types. The impacts of CMA on mobile computing are
presented in Section III. Section IV presents the analysis
and taxonomy of varied cloud-based augmentation resources.
Comprehensive survey of the state-of-the-art CMA approaches
is presented and taxonomy is devised in Section V. We
discuss the CMA decision making and limitation factors and
illustrate CMA feasibility in Section VI. Finally, open research
challenges are presented in Section VII and paper is concluded
in Section VIII.
II. MOBILE COMPUTATION AUGMENTATION
In this Section, we present a de?nition on mobile computing
augmentation based on the available de?nitions on the relevant
concepts, particularly remote execution [5] and cyber foraging
[6]. Additionally, the motivation for performing mobile com-
putation augmentation is described and taxonomy of mobile
augmentation types is presented.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 4
TABLE II
INITIAL FEATURES OF MOBILE EMPOWERMENT APPROACHES.
Approach Architecture Client Load Migration Partitioning Server Mobility
Load Sharing Client-Server Entire Task Entire task NA Server NA
Remote
Client-Server Entire Task Entire/partial Static
Server No
Execution /desktop
Cyber Client-Server
Entire Task Entire/partial Dynamic
Surrogates No
Foraging Peer-to-Peer
Mobile Varies, e.g.,
/Nil
Entire/partial/ Static & Server, Yes
Computation Client-server Entire/partial Nil migration dynamic surrogate
Augmentation P2P, Adhoc (Use remote &mobile
collaborative services)
A. De?nition
Indeed, empowering computation capabilities of mobile
devices is not a new concept and there have been different
approaches to achieve this goal, including load sharing [4],
remote execution [5], cyber foraging [6], and computation
of?oading [60], [61] that are described as follows. We have
analyzed them and summarized the analysis results in Table
II. Results in this Table are extracted from the early efforts in
each category, which are already deviated from their original
characteristics due to the research achievements.
• Load Sharing: Othman and Hailes’ work [4] in 1998
can be considered as one of the earliest efforts to conserve
native resources of mobile devices using a software approach.
The main idea is inspired from the concept of load balancing
in distributed computing that is “a strategy which attempts
to share loads in a distributed system without attempting
to equalize its load” [4]. This approach migrates the whole
computation job for remote execution. It considers several
metrics such as job size, available bandwidth, and result size
to identify if the load balancing and transferring the job to
the remote computer can save energy. However, they need to
send the task and data to the nearest base station and wait for
the results to return. The base station is responsible to ?nd
appropriate server to run the job and forward the results back
to the mobile device. Moreover, computing server is a ?xed
computer and there is no provision for user and code mobility
at run time.
• Remote Execution: The concept of remote execution for
mobile clients emerged in 90s and several researchers [5],
[62]–[65] endeavor to enable mobile computers to performing
remote computation and data storage to conserve their scarce
native resources and battery. In 1998 [5], feasibility of remote
execution concept on mobile computers, particularly laptops
was investigated. The authors report that remote execution can
save energy if the remote processing cost is lower than local
execution. Remote execution involves migrating computing
tasks from the mobile device to the server prior the execution.
The server performs the task and sends back the results to
the mobile device. However, difference between computation
power of client and server is not a metric of decision making
in this method. Moreover, the whole task needs to be migrated
to the remote server prior the execution which is an expensive
effort. It also neglects the impact of environment characteris-
tics on the remote execution outcome. Static decision making
is another shortcoming of this proposal.
• Cyber Foraging: Satyanarayana in 2001 [6] further
developed the remote execution idea by considering dynamism
in remote execution process. The author de?ned cyber forag-
ing as the process “to dynamically augment the computing
resources of a wireless mobile computer by exploiting wired
hardware infrastructure”. Resources in cyber foraging are
stationary computers or servers in public places —connected
to wired Internet and power cable—that are willing to perform
intensive computation on behalf of the resource-constraint
mobile devices in vicinity.
However, load sharing, remote execution, and cyber forag-
ing approaches assume that the whole computing task is stored
in the device and hence, it requires the intensive code and data
to be identi?ed and partitioned for of?oading —either stati-
cally prior the execution or dynamically at runtime —which
impose large overhead on resource-poor mobile device [19].
Moreover, as Kumar et al. [66] explain, for each mobile
user that runs the intensive application, the whole of?oading
process must be repeated including decision making process
in the device and transferring the heavy components and large
data to the network. Due to slight differences among these
concepts, researchers use the terms ‘remote execution’, ‘cyber
foraging’, and ‘computation of?oading’ interchangeably in the
literature with similar principle and notion.
Nevertheless, researchers in recent activities [36], [42], [45],
[46] aim to enhance computing capabilities of mobile devices
in a slightly different manner. They assume to store the
intensive code and data outside the device and keep the rest in
the mobile device instead of storing the whole task —including
both lightweight and intensive code and data —in the mobile
device. Therefore, the overhead of identifying, partitioning,
and migrating the resource-intensive task is mitigated, energy
is saved, and storage problem is alleviated in mobile devices.
Moreover, storing intensive components outside the device,
in a publicly accessible storage, can increase their reusability
and enable more than one user to leverage their computation
services. Therefore, we coin the term mobile computation
augmentation as the wider phrase that subsumes load sharing,
remote execution, cyber foraging, and other approaches that
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 5
augment computing capabilities of mobile devices.
• Mobile Computation Augmentation: Mobile computa-
tion augmentation, or augmentation in brief, is the process of
increasing, enhancing, and optimizing computing capabilities
of mobile devices by leveraging varied feasible approaches,
hardware and software. Mobile device is any non-stationary,
battery-operating computing entity able to interact with end-
user and execute transactions, store data, and communicate
with the environment using wireless technologies and varied
sensors. Smartphone, Tablet, handheld/wearable computing
devices, and vehicle mount computers are mobile device in-
stances. Approaches that can augment mobile devices include
hardware and software. Hardware approach involves manu-
facturing high-end physical components, particularly CPU,
memory, storage, and battery. Software approaches can be
—but are not limited to —computation of?oading, remote data
storage, wireless communication, resource-aware computing,
?delity adaptation, and remote service request (e.g., context
acquisition).
Augmentation approaches can increase computing capabil-
ities of mobile devices and conserve energy. They can be
leveraged in three main categories of applications, namely (i)
computing-intensive software such as speech recognition and
natural language processing, (ii) data-intensive programs such
as enterprise applications, and (iii) communication-intensive
applications such as online video streaming applications. The
augmented mobile device is able to perform complex tasks that
could not otherwise perform. Hence, the mobile application
developers do not take into account resource shortcomings
of mobile devices in developing mobile application and users
will not consider their devices weaknesses in utilizing varied
complex applications.
B. Motivation
Mobile devices have recently gained momentous ground in
several communities like governmental agencies, enterprises,
social service providers (e.g., insurance, Police, ?re depart-
ments), healthcare, education, and engineering organizations
[67], [68]. However, despite of signi?cant improvement in
mobiles’ computing capabilities, still computing requirements
of mobile users, especially enterprise users, is not achieved.
Several intrinsic de?ciencies of mobile devices encumber
feasibility of intense mobile computing and motivate aug-
mentation. Leveraging augmentation approaches, vision of
performing intense mobile operations and control such as
remote surgery, on-site engineering, and visionary scenarios
similar to the lost child and disaster relief described in [69]
will become reality. In this part, we analyze and taxonomize
smartphones’ de?cits that can be alleviated by augmentation.
Figure 2 depicts our devised taxonomy.
1) Processing Power: Processing de?ciencies of mobile
clients due to slow processing speed and limited RAM is one
of the major challenges in mobile computing [69]. Mobile de-
vices are expected to have high processing capabilities similar
to computing capabilities of desktop machines for performing
computing-intensive tasks to enrich user experience. Realizing
such vision requires powerful processor being able to perform
large number of transactions in a short course of time.
Large internal memory/RAM to store state stack of all
running applications and background services is also lacking.
Beside local memory limitations, memory leakage also inten-
si?es memory restrains of mobile devices. Memory leakage
is the state of memory cells being unnecessarily occupied by
running applications and services or those cells that are not
released after utilization. Garbage-collector-based languages
like Java in Android
2
contribute to memory leakage due to
failed or delayed removal of unused objects from the memory
[70]. Android’s kernel level transactions can also leak memory
in the absence of memory management mechanisms [70],
[71]. Moreover, inward de?ciency and inef?cient design and
implementation of mobile applications can also waste scarce
memory of mobile devices. Thus, in the absence of required
memory, applications are frequently paused or terminated
by the operating system leading to longer execution time,
excessive resource dissipation, and ultimately mobile user
experience degradation.
2) Energy Resources: Energy is the only non-replenishable
resource in mobile devices that demands external resources
to be replenished [72], [73]. Currently, energy requirement of
a mobile device is supplied via lithium-ion battery that lasts
only few hours if device is computationally engaged. Battery
capacity is increasing at about 5 to 10% a year [74], [75]
as battery cells are excessively dense [72]. Moreover, mo-
bile device manufacturers endeavor to attain device lightness,
compactness, and handiness, which prevent exploitation of
bulky long-lasting batteries. User safety is another concern that
con?nes manufacturers to produce low capacity batteries [76].
While explosion of a battery with few hundreds milliamperes
capacity can jeopardize human life [77], explosion of a high-
capacity battery can carry catastrophic consequences.
Energy harvesting efforts [78]–[80] seek to replenish energy
from renewable resources, particularly human movement, solar
energy, and wireless radiation, but these resources are mostly
intermittent and not available on-demand [81]. For instance,
a sitting mobile user at night cannot have any external power
source in the absence of wall power and wireless radiations.
Moreover, researchers aim at reducing the energy overhead
in different aspects of computing, including hardware, OS,
application, and networking interface [82], [83]. Efforts are di-
rected to develop alternative energy resources such as nuclear
batteries that will likely last months or years [84]. However,
signi?cant deal of R&D is needed to ful?ll ever-increasing
energy requirements of mobile users.
Hence, in the absence of long-spanning energy on mobile
devices, alternative augmentation approaches play a vital role
in maturing mobile and ubiquitous computing.
3) Local Storage: Drastic increasing in the number of
applications and amount of digital contents such as pictures,
songs, movies, and home ?lms [85] from one hand and limited
storage of mobile devices from the other hand decelerate
usability of mobile devices. While PCs are able to locally
store huge amount of data, smartphones are limited to few
gigabytes of space which are mostly occupied by system
?les, user applications, and personal data. Therefore, frequent
2
urlhttp://www.android.com/
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 6
Fig. 2. Taxonomy of Augmentation Motivation: Intrinsic and non-intrinsic mobile challenges motivate augmentation.
storing, updating, and deleting data as well as uninstalling and
reinstalling applications due to space limitation cause irksome
impediments to mobile users [86]. Additionally, delivering
of?ine usability, which is one of the most important character-
istics of contemporary applications, remains an open challenge
since mobile devices lack large local storage.
4) Visualization Capabilities: Effective data visualization
on small mobile devices’ screen is a non-trivial task when
current screen manufacturing technologies and energy limita-
tions do not allow signi?cant size extensions without losing
device handiness. Currently smartphones like HTC One X
3
and Samsung Galaxy Note II
4
have the biggest screens, at
4.7 and 5.5 inches respectively; however, they are very small
compared to PCs and notebooks.
Therefore, ef?cient data visualization in small smartphones’
screen necessitates software-based techniques similar to tab-
ular pages, 3D objects, multiple desktops, switching between
landscape and portrait views (needs accelerometer), and verbal
communication to virtually increase presentation area. Re-
cently, computing-intensive mobile 3D display technology is
promising to noticeably mitigate the visualization de?cit of
contemporary smartphones. Glass-free auto-stereoscopic dis-
plays [87] can present 3D data by exploiting binocular parallax
to offer a different view for each eye. Taking advantages
of current and imminent software-based techniques beside
native tools, especially tilting sensors signi?cantly improve the
mobile visualization capabilities in the near future. However,
these approaches are computation-intensive processes that
quickly drain battery [87], [88]. A feasible alternative solution
to realize software-based content presentation approaches is to
augment smartphones’ computing capabilities.
5) Security, Privacy, and Data Safety: Mobile end-users
are concerned about security and privacy of their personal
data, banking records, and online behaviors [89]. The dramatic
increase in cybercrime and security threats within mobile
devices [90], cloud resources [91] and wireless transactions
makes security and privacy more challenging than ever [92].
Moreover, performing complex cryptographic algorithms is
likely infeasible because of computing de?ciencies of mobile
3http://www.htc.com/www/smartphones/htc-one-x/
4http://www.samsung.com/my/consumer/mobile-devices/galaxy-
note/galaxy-note/GT-N7100RWDXME
devices. Securing ?les using pair of credentials is also less
realistic in the absence of large keyboard.
Data safety is another challenge of mobile devices, because
information stored inside the local storage of mobile devices
are susceptible to safety breaches due to high probability of
hardware malfunction, physical damage, stealing, and loss.
Amalgam of these problems and de?ciencies in mobile
computing stimulates researchers from academia and industry
to exploit novel technologies and approaches to augment
computing capabilities of mobile devices which is subject of
this study.
C. Mobile Augmentation Types: Taxonomy
In this Section, we analyze and classify augmentation ap-
proaches into two major types of hardware and software. Our
devised taxonomy is depicted in Figure 3 and described as
follows.
Hardware. The hardware approach aims to empower smart-
phones by exploiting powerful resources, particularly multi-
core CPUs with high clock speed [93], large screen, and long-
lasting battery [84], [94]. ARM
5
and Samsung
6
are major mo-
bile processor manufacturers producing multi-core processors
such as ARM Cortex-59
7
and Samsung Exynos 5 Octa core
8
that perform in higher speed than a single core processors
[93]. However, doubling the CPU clock speed approximately
octuples the device energy consumption [66].
Nevertheless, augmentation via sophisticated hardware is
hindered by several obstacles. Firstly, generating powerful
processor, large storage, and big screen decrease smartphone
handiness due to additional heat, size, and weight. Secondly,
considering the fact that utilizing long-lasting battery in small
mobile devices is not feasible with current technologies, re-
source enlargement contributes toward faster battery drainage
and shorter battery life time. Thirdly, equipping mobile de-
vices with high-end hardware noticeably increases their price
5http://arm.com
6http://samsung.com
7http://www.arm.com/products/processors/cortex-a50/cortex-a57-
processor.php
8http://www.samsung.com/global/business/semiconductor/minisite/Exynos/
blog CES 2013 Samsung Mobilizes Possibility with Exynos5Octa.html
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 7
Fig. 3. Taxonomy of Mobile Augmentation Types.
compare to the stationary machines. Unlike PCs, smartphone’s
hardware is not upgradable; hence, a new device should
be possessed in case of technology advancement. Therefore,
in the absence of futuristic engineering technologies, the
hardware-based augmentation process is slow and expensive
that necessitates alternative augmentation approaches to en-
hance computing capabilities of mobile devices without drastic
ownership price hike.
Software. Software-oriented mobile augmentation approaches
are classi?ed into ?ve groups and will be explained later in
this part. Resources that are used in major software-oriented
approaches are classi?ed into two groups, namely traditional
and cloud-based. Their major differences lie on resource pro-
visioning and access strategies, service security and delivery
models, and resource characteristics. In traditional approaches,
researchers leverage centralized resources of distant traditional
servers or free nearby surrogates. Several problems such
as resource availability, elasticity, and security of traditional
approaches hinder their success. For instance, surrogates can
terminate their services anytime without considering their
current load, and can violate user security and privacy by
changing execution sequence or altering raw and processed
data.
To alleviate the problems of traditional servers, researchers
in recent efforts [25], [27], [29], [31], [33]–[35], [41], [43],
[44], [95] exploit highly available, elastic, secure cloud in-
frastructure. “Cloud is a type of parallel and distributed
system consisting of a collection of interconnected and vir-
tualized computers dynamically provisioned and presented as
one or more uni?ed computing resources based on service-
level agreements established through negotiation between the
service provider and consumers” [20].
While utilizing cloud resources, users pay for the amount
and duration they utilize various resources (e.g., CPU, mem-
ory, and bandwidth) based on an agreed SLA. In the SLA, the
amount and quality of required resources such as processor,
RAM, and storage are speci?ed and user is billed accordingly.
Service delivery failure will be compensated by the vendor.
Lucrative ?nancial bene?ts of cloud services encourage cloud
providers to compete in delivering high service availability,
reliability, security, and robustness to increase their market
share. Hence, the augmentation performance is less affected
by resource unreliability and interruption.
Moreover, cloud infrastructures are available to end-users
via Virtual Machine(VM)
9
to increase resource utilization
ratio and enhance overall security and privacy. Virtualization
technology aims to enable resource sharing in an isolated
environment called VM. It realizes execution of multiple
operating systems on a single machine and enables sharing
of large resources among multiple end-users. Users can only
access to infrastructures allocated to their VMs and cannot
access prohibited resources and contents.
Table III summarizes the comparison results of traditional
and cloud-based resources and advocates differentiations be-
tween the conventional servers and clouds. High computing
power, elasticity, mobility support, low utilization overhead,
and security are some of the signi?cant advantages of cloud
resources compare to the surrogates that advocate the latest
paradigm shift in mobile augmentation.
Software augmentation techniques are classi?ed as remote
execution (of?oading or cyber foraging) [5]–[8], [10]–[13],
[16]–[18], [25], [29], [30], [33]–[35], [41], [43], [44], [96],
remote storage [97], Multi-tier programming [36], [45], [46],
live cloud-streaming [98], resource-aware computing [99],
[100], and ?delity adaptation [101] and explained as follows.
• Remote execution: As explained in II-A,the resource-
hungry components of mobile applications —in whole or
part —are migrated to the resource-rich computing device(s)
that are willing to share their resources with mobile devices.
Rapid development of heterogeneous mobile devices obliges
adaptive of?oading approaches able to enhance capabilities
of wide range of mobile devices in dynamic environment
with least processing overhead and latency. The ef?ciency of
of?oading approaches highly depends on what component(s)
can be partitioned? When partitioning takes place? Where to
execute the component(s)? And how to communicate with the
remote server? [102]. Of?oading approaches perform varied-
time analysis to answer these questions, which are classi?ed
into three groups and explained as below.
Design Time Analysis: In this method, the application’s
complexity is analyzed at design time to determine the answer
of four above questions. Application developer or a middle-
ware speci?es the resource-intensive components of mobile
9http://www.vmware.com/virtualization/what-is-virtualization.html
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 8
TABLE III
COMPARISON BETWEEN TRADITIONAL AND CLOUD-BASED COMPUTING RESOURCE.
Features Traditional Cloud-Based
Computation Power Low High
Elasticity Low High
User Experience Low High
Reliability Low High
Availability Intermittent On-demand
Client Mobility Limited Unlimited
Multi-tenancy Not available Available
Serving Incentive Not provisioned Provisioned
Utilization Cost Free Pay-As-You-Use
Utilization Overhead High Low
Management Decentralized De/Centralized
Back-end Connectivity Wired Wired & Wireless
Communication Latency Low Varied
Computation Latency High Low
Security Low High
Data Safety Low High
application that can be of?oaded to the remote server and label
them as remote component(s). Programmers decide how to
partition application and adapt its performance to the dynamic
mobile environment which is a non-trivial task, mainly due
to the lack of knowledge about the execution environment.
Performing such action needs excessive programming skill
and knowledge of computation of?oading. Design time ap-
proaches [8], [10], [12] notably save native resources of
mobile device by reducing the processing and monitoring
overheads. However, partitioning prior to the execution is not
always optimal and cannot accurately adapt performance in
diverse execution environments and also imposes extra efforts
on the application developer or middleware for deciding on
partitioning. Hence, design time partitioning approaches are
likely become obsolete.
Runtime Analysis: Runtime or dynamic partitioning referred
to methods such as [25], [103] that aims to answer four
questions at runtime. They identify and partition the resource-
hungry parts of the application, specify how and where to
execute the partitioned components [102], [104], and de-
termine how to communicate with the server. In dynamic
methods, resource requirement of the application is analyzed
and available resources are detected to decide if the appli-
cation requires remote resources. Upon decision making the
system performs of?oading. Further monitoring is necessary
to gather knowledge of available remote resources to maintain
execution history. Although these approaches provide dynamic
and ?exible solutions, large amount of resources are wasted
at runtime that prolongs application execution and decrease
energy ef?ciency.
Hybrid Analysis: The ultimate aim of hybrid approaches
[105] is to increase performance and ef?ciency of augmen-
tation methods. Deciding on how to perform the of?oading
mainly depends on the native resources, remote resources, and
available network bandwidth. In [105], prior to the application
execution, the system decides based on four options, namely i)
no action, ii) dynamic, iii) static, and iv) pro?le only whether
to of?oad or not and in case of of?oading specify what type
of partitioning should take place. The pro?le only option is
similar to the no action, but the systems collect execution
information to maintain execution history for future purpose.
• Remote Storage: Remote storage is the process of ex-
panding storage capability of mobile devices using remote
storage resources. It enables maintaining applications and
data outside the mobile devices and provides remote access
to them. In early efforts, researchers in [97] utilize iSCSI
(Internet Small Computer System Interface) [106] —as a
well-established protocol for remote storage —to access the
server’s I/O resources via mobile clients over the TCP/IP
network to store, backup, and mirror data [107]. However,
the throughput of iSCSI is highly affected by the mobile-
server distance. Using iSCSI is also dif?cult for handling
large ?les such as multimedia and database ?les. Moreover,
due to message passing in wireless medium through TCP/IP,
the security and processing overhead (e.g., cryptography and
data compression) are further challenges. To alleviate these
challenges, several researches as MiSC [108], UbiqStor [109],
[110], and Intermediate Target [111] are proposed towards
realizing remote storage on mobile devices. However, due to
scalability, availability, performance, and ef?ciency issues of
traditional servers, power of remote storage could not fully
unleash using traditional servers.
Several proposals and data storage services in academia
and industry aim to expand mobile storage by exploiting
cloud computing, especially Jupiter [31], SmartBox [112],
Amazon S3
10
, Mozy
11
, Google Docs
12
, and DropBox
13
. For
instance, Jupiter expands smartphone storage and assists end-
users in organizing large applications and data. Jupiter lever-
10http://aws.amazon.com/s3/
11http://mozy.com
12https://docs.google.com/
13https://www.dropbox.com/
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 9
ages cloud infrastructures to store big data of mobile users.
Heavy applications are executed inside the cloud’s VM of
smartphones and results are forwarded to the physical device
after execution. Amazon S3 is a general purpose storage
offering simple operations to store and retrieve cloud data
while Mozy provides data backup facilities with main focus
on enhancing cloud data safety against natural disasters.
• Resource-Aware Computing: In resource-aware comput-
ing efforts, especially [99], [100], [113]–[115], resource re-
quirements of mobile applications are diminished utilizing
the application-level resource management methods (using
application management software such as compiler and OS)
and lightweight protocols. Resource conservation is performed
via ef?cient selection of available execution approaches
and technologies [114]. Any mobile application consists of
application-level resource management method is considered
as a resource-aware application. For instance, in [100], authors
propose an energy-friendly scheme for content-based image
retrieval applications using three of?oading options, namely
i) local extraction-remote search, ii) remote extraction-remote
search, and iii) remote extraction-local search. The authors
consider available bandwidth, image database size, and num-
ber of user queries to opt any of three of?oading options
for saving energy. In a high bandwidth network with limited
queries, the third option is bene?cial; the system uploads all
un-indexed images to the remote server and receives the results
to be loaded into the memory. Then, all search queries are
executed locally.
Similarly, applications can decide whether to choose 2G or
3G in telephony and FTP. Using 2G network for telephony and
3G for FTP can noticeably reduce resource requirements of
the mobile applications, according to the power consumption
patterns presented in [116]. 2G network technology consumes
less energy for establishing a telephony communication, while
3G is more energy-friendly for ?le transfer transactions.
• Fidelity adaptation: Fidelity adaptation is an alternative
solution to augment mobile devices in the absence of remote
resources and online connectivity. In this method local re-
sources are conserved by decreasing quality of application
execution, which is unlikely desirable to end-users. As a
well-known ?delity adaptation approach, we can refer to
the YouTube
14
. Users in YouTube can adjust the streaming
quality based on available bandwidth. To achieve optimized
performance, researchers [78], [117] leverage composition of
cyber foraging and ?delity adaptation.
• Multi-tier Programming: Developing distributed multi-
tier mobile applications leveraging remote infrastructures is
another technique employed in efforts such as [36], [45], [46],
[118] to reduce resource requirements of mobile applications.
The main idea in this type of mobile applications is to
reduce the client-side computing workload and develop the
applications with less native resource requirements. Certainly,
the computationally intensive components of the applications
are executed outside the device, whereas the interactive (user
interface) and native codes (e.g., accessing to the device
camera) remain inside the device for execution.
14http://youtube.com
Multi-tier applications are lightweight aiming to consume
the least possible local resources by utilizing remote compo-
nents and services, whereas native applications are monolithic
applications often require runtime migration for execution.
Therefore, monitoring time and communication overhead of
multi-tier applications are shrunk leading to explicit resource
saving and user experience enhancement.
• Live Cloud Streaming: In recent efforts to harness cloud re-
sources, researchers from Onlive
15
and Gaikai
16
, among other
organizations introduce new approach to augment computing
capabilities of mobile devices, entitled live cloud streaming
[98]. In live cloud streaming approaches, mobile device acts
as a dump client able to interact with server using a browser
or application GUI. In live cloud streaming applications, entire
processing take place in the cloud and results are streaming
to the mobile devices. However, usability of cloud-streaming
is hindered by latency, network bandwidth, portability, and
network traf?c cost.
Functionality of cloud-streaming applications absolutely de-
pends on the network availability and the Internet. Transferring
mobile-user input to the server is another critical factor that
requires considerable attention under wireless Internet connec-
tion. Moreover, since majority of mobile network providers
deploy ‘pay-as-you-use’ data plans, the large data traf?c of
cloud-streaming services imposes high communication cost
on users. Yet congestion handling remains an open issue at
peak hours. Entirely relying on cloud-streaming infrastructures
and avoiding smartphones resources’ utilization impact on
application responsiveness and levy extravagant ownership,
maintenance, power, and networking expenses to the cloud-
streaming service providers, which is not a green computing
approach.
III. IMPACTS OF CMA ON MOBILE COMPUTING
This Section discusses the advantages and disadvantages
of performing a CMA process on mobile computing that
are summarized in Table IV. We aim to demonstrate how
CMA approaches mitigate de?ciencies of mobile computing
explained in Section II-B. In this Section the terms ‘cloud
resources’ and ‘cloud infrastructures’ refer to any type of
cloud-based resources and infrastructures discussed in Section
V.
A. Advantages
In this part, eight major bene?ts of utilizing cloud resources
in mobile augmentation processes are introduced.
1) Empowered Processing: Empowering processing is the
state of virtually increased transaction execution per second
and extended main memory leveraging CMA approaches. In
computing-intensive mobile applications, either the hosting
device does not have enough processing power and memory
or cannot provide required energy. A common solution is to
of?oad the application —in whole or part —to a reliable,
powerful resource with least energy and time cost. In compu-
tation of?oading, the complex, CPU- and memory-intensive
15http://onlive.com
16http://gaikai.com
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 10
components of a standalone application are migrated to the
cloud. Consequently, the mobile devices can virtually perform
and actually deliver the results of heavy transactions beyond
their native capabilities.
Although surrogates in traditional augmentation approaches
[8]–[10], [12] could increase computing capabilities of mobile
devices, excessive overhead of arbitrary service interruption
and denial could shadow augmentation bene?ts [19], [119].
Cloud resources guarantee highest possible resource availabil-
ity and reliability.
Leveraging CMA approaches, application developers build
mobile application with no consideration on available native
resources of mobile devices and mobile users dismiss their
devices’ inabilities. Hence, computing- and memory-intensive
mobile applications like content-based image retrieval appli-
cations (enable mobile users to retrieve an image from the
database) can be executed on smartphones without excess
efforts.
However, a ?exible and generic CMA approach that can
enhance plethora of mobile devices with least con?guration,
processing overhead, and latency is a vital need in excessively
diverse mobile computing domain. Such diversity is mainly
due to the rapid development of smartphones and Tablets,
and sharp rise in their hardware, platform, API, feature, and
network heterogeneity [120] in the absence of early standard-
ization.
2) Prolonged Battery: Long-lasting battery can be con-
sidered as one the most signi?cant achievements of CMA
approaches for large number of mobile users. Smartphone
manufacturers have already utilized high speed, multi-core
ARM processors (e.g., Cortex-A57 Processor
17
) being able to
perform daily computing needs of mobile end-users. How-
ever, such giant processing entities consume large energy
and quickly drain the battery that irks end-users. CMA so-
lutions can noticeably save energy [95] by migrating heavy
and energy-intensive computing to the cloud for execution.
Although energy ef?ciency is one of the most important
challenges of current CMA systems, several efforts such as
[53]–[55], [121], [122] are endeavoring to comprehend the
energy implications of exploiting cloud-based resources from
mobile devices and shrinking their energy overhead.
In traditional cyber foraging or surrogate computing ap-
proaches, energy is saved by computation of?oading, but
several issues such as lack of mobility support and resource
elasticity can neutralize the bene?ts of energy-hungry task
of?oading.
3) Expanded Storage: In?nite cloud storage accessible
from smartphones enables users to utilize large number of
applications and digital data on device. Hence, they are not
obliged to frequently install and remove popular applications
and data due to the space limit. Online connectivity is essential
to access cloud storage. In such online storage systems, data
are manually or automatically updated to the online storage
for maintaining the consistency of the online storage system.
Storing applications in cloud storage provides the opportu-
nity to update the code without consuming any mobile I/O
17http://www.arm.com/products/processors/cortex-a50/cortex-a57-
processor.php
transactions which enhances user experience and improves the
smartphones’ energy ef?ciency —because I/O transactions are
energy-hungry tasks.
4) Increased Data Safety: CMA efforts can bring the
bene?t of data safety to the mobile users. Naturally, stored
data on mobile devices are susceptible to loss, robbery,
physical damage, and device malfunction. Storing sensitive
and personal data such as online banking information, online
credentials, and customer related information on such a risky
storage signi?cantly degrades the quality of user experience
and hinders usability of mobile devices. Due to the scarce
computing resources, especially energy in mobile devices,
performing complex and secure encryption provisions is not
feasible. Hence, by storing data in a reliable cloud storage
[112], [123], users ensure data availability and safety regard-
less of time, place, and unforeseen mishaps. Threats such as
device robbery or physical damage to the mobile devices will
effect on the tangible value of the device rather than intangible
value of the data.
5) Ubiquitous Data Access and Content Sharing: Cloud
infrastructures play a vital role in enhancing data access
quality. Storing data in cloud resources enables mobile users
to access their digital contents anytime, anywhere, from any
device. Hence, the impact of temporal, geographical, and
physical differences is noticeably decreased that enriches user
experience.
Moreover, cloud storages facilitate data sharing and contri-
bution among authorized users. Every ?le and folder in cloud,
usually has a protected unique access link that can be obtained
by the owner to share them among legitimate users. Network
traf?c is hence, shrunk because data is accumulated in a central
server accessible to unlimited users from various machines.
Cloud can signi?cantly enhance data transfer among different
mobile devices. One of the most irksome user’s impediments
is to transfer data from current mobile device to a new handset.
Apart from its temporal cost, porting data from one device to
another, especially to a heterogeneous device is a risky practice
that puts data is in the risk of corruption and loss of integrity.
Stored data on Cloud remain safe and can be synchronized to
any number of mobile devices with minimum risk. However,
a reliable data access control mechanism is required to adjust
user permissions.
6) Protected Of?oaded Content: Cloud storage solutions
aim to protect remote codes and data while ensure user’s
privacy. This is one of the most important gains of replacing
surrogates with cloud resources. Cloud servers deploy virtu-
alization technology to isolate the guest environment from
other guests and also from their permanent software stack.
Moreover, cloud vendors deploy strict security and privacy
policies to not only ensure con?dentiality of user content,
but also to protect their properties and business. Implement
internal security provisions particularly the state-of-the-art bio-
metric security systems to protect their physical infrastructures
and avoid unauthorized access. Employing complex content
encryption, frequent patching, and continuous virus signature
update inside the company premise or seeking technical ser-
vices from a trusted third party [124] are other examples of
security provisions undertaken in cloud to further protect cloud
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 11
TABLE IV
IMPACT OF CMA APPROACHES IN MOBILE COMPUTING.
Advantages Disadvantages
Empowered Processing Dependency to High Performance Networking Infrastructure
Prolonged Battery Excessive Communication Overhead and Traf?c
Expanded Storage Unauthorized Access to of?oaded Data
Increased Data Safety Application Development Complexity
Ubiquitous Data Access Paid Infrastructures
and Content Sharing
Protected Of?oaded Contents Inconsistent Cloud Policies and Restrictions
Enriched User Interface Service Negotiation and Control
Enhanced Application Generation Nil
storage.
7) Enriched User Interface: As described in II-B4, vi-
sualization shortcomings of mobile devices diminish user
experience and hinder smartphones’ usability. However, cloud
resources can be exploited to perform intensive 2D or 3D
screen rendering. The ?nal screen image can be prepared based
on the smartphone screen size and streamed to the device.
Consequently, screen adaptation also is achieved when cloud
side processing engine automatically alter the presentation
technique to match screen image with the device screen size.
8) Enhanced Application Generation: Cloud resources and
cloud-based application development frameworks similar to
µCloud and CMH, facilitate application generations in het-
erogeneous mobile environment. Once a cloud component is
built, it can be utilized to develop various distributed mobile
applications for large number of dissimilar mobile devices. In
the presence of cloud components, application programmer
needs to develop native mobile components and integrate
them with relevant, prefabricated cloud components to develop
a complex application. When a mobile-cloud application is
developed for Android device, by slightly changing native
components the application is transited to new OS like iOS
18
and Symbian
19
which signi?cantly save time and money.
B. Disadvantages
Despite of many advantageous aspects of cloud services,
their success is hindered by several drawbacks and shortcom-
ings that are discussed as follows.
1) Dependency to High Performance Networking Infras-
tructure: CMA approaches demand converged wired and
wireless networking infrastructures and technologies to ful?ll
intersystem communication requirements. In wireless domain,
CMAs need high performance, robust, reliable, high band-
width wireless communication to realize the vision of com-
puting anywhere, anytime, from any-device. In wired commu-
nication, fast reliable communications ground is essential to
facilitate live migration of heavy data and computations to a
regional cloud-based resources near the mobile users. Efforts
such as next generation wireless networks [125] and the open
mobile infrastructure [126] with Open Wireless Architecture
18http://www.apple.com/ios/
19http://licensing.symbian.org/
(OWA) by Sieneon [127] contribute toward enhancing the
networking infrastructures’ performance in MCC.
2) Excessive Communication Overhead and Traf?c: Mobile
data traf?c is signi?cantly growing by ever-increasing mo-
bile user demands for exploiting cloud-based computational
resources. Data storage/retrieval, application of?oading, and
live VM migration are example of CMA operations that
drastically increase traf?c leading to excessive congestion
and packet loss. Thus, managing such overwhelming traf?c
and congestion via wireless medium becomes challenging,
especially when of?oading mobile data are distributed among
helping nodes to commute to/from the cloud. Consequently,
application functionality and performance decrease leading to
user experience degradation.
3) Unauthorized Access to Of?oaded Data: Since cloud
clients have no control over their remote data, users contents
are in risk of being accessed and altered by unauthorized
parties. Migrating sensitive codes as well as ?nancial and
enterprise data to publicly accessible cloud resources decreases
users privacy, especially enterprise users. Moreover, storing
business data in the cloud is likely increasing the chance
of leakage to the competitor ?rm. Hence, users, especially
enterprise users hesitate to leverage cloud services to augment
their smartphones.
4) Application Development Complexity: The excessive
complexity created by the heterogeneous cloud environment
increases environment’s dynamism and complicates mobile
application development. Mobile application developers are
required to acquire extensive knowledge of cloud platforms
(i.e., cloud OSs, programming languages, and data structures)
to integrate cloud infrastructures to the plethora of mobile
devices. Understanding and alleviating such complexity im-
pose temporal and ?nancial costs on application developers
and decrease success of CMA-based mobile applications.
5) Paid Infrastructures: Unlike the free surrogate resources,
utilizing cloud infrastructure levies ?nancial charges to the
end-users. Mobile users pay for consumed infrastructures
according to the SLA negotiated with cloud vendor. In certain
scenarios, users prefer local execution or application termina-
tion because of monetary cloud infrastructures cost. However,
user payment is an incentive for cloud vendors to maintain
their services and deliver reliable, robust, and secure services
to the mobile users.
In addition, cloud vendors often charge mobile users twice;
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 12
Fig. 4. Taxonomy of Cloud-based Computing Resources.
once for of?oading contents to the cloud and once again when
users decide to transfer their cloud data to another cloud
vendors to utilize more appropriate service (e.g., monetary and
QoS (Quality of Service) aspects).
6) Inconsistent Cloud Policies and Restrictions: One of the
challenges in utilizing cloud resources for augmenting mobile
devices is the possibility of changes in policies and restrictions
imposed by the cloud vendors. Cloud service providers apply
certain policies to restrain service quality to a desired level by
applying speci?c limitations via their intermediate applications
like Google App Engine bulk loader
20
. Services are controlled
and balanced while accurate bills will be provided based on
utilized resources.
Also, service provisioning, controlling, balancing, and
billing are often matched with the requirements of desktop
clients rather than mobile users [128]. Considering the great
differences in wired and wireless communications, disregard-
ing mobility and resource limitations of mobile users in
design and maintenance of cloud can signi?cantly impact on
feasibility of CMA approaches. Hence, it is essential to amend
restriction rules and policies to meet MCC users requirements
and realize intense mobile computing on the go.
7) Service Negotiation and Control: While cloud users are
required to negotiate and comply with the cloud terms and
conditions for a certain period of time, often cloud agreements
are nonnegotiable and policies might change over the time.
Moreover, there is no control over the cloud performance and
commitments in the absence of a controlling authority or a
trusted third party. Hence, CMA services are always volatile
to the service quality of cloud vendors.
IV. TAXONOMY OF CLOUD-BASED COMPUTING
RESOURCES
Researchers [24]–[27], [27]–[43], [45]–[49] aim to obtain
user requirements and preferences by exploiting varied types
20https://developers.google.com/appengine/docs/python/tools/uploading
data
of cloud-based resources to augment computing capabilities
of resource-constraint smartphones. Based on the distance
and mobility traits of such varied cloud-based computing
resources, we classify them into four groups, namely distant
immobile clouds, proximate immobile computing entities,
proximate mobile computing entities, and hybrid that are
taxonomized in Figure 4 and explained as follows. Table V
represents the comparison results of these cloud-based com-
puting resources. This Table can be utilized as a guideline for
appropriate selection of cloud-based infrastructures in future
CMA researches.
A. Distant Immobile Clouds
Public and private clouds comprised of large number of
stationary servers located in vendors or enterprises premises
are classi?ed in this category. They are highly available,
scalable, and elastic resources that are often located far from
the mobile nodes accessible via the Internet. Although public
cloud resources are likely more secure compared to the other
types of resources due to complex security provisions and
on-premise infrastructures [129]–[132], they are vulnerable
to security attacks and breaches like Amazon EC2 crash
[92] and Microsoft Azure security glitch [133]. Accessing
cloud resources, especially public clouds often carries the
risk of communicating through the risky channel of Internet.
However, giant clouds are endeavoring to maintain security
-for more market share- and could establish high reputation-
based trust by providing long-term services to the users.
Additionally, the performance and ef?cacy of these ap-
proaches are affected by long WAN latency due to the long
distance between mobile client and stationary cloud data cen-
ters. One potential approach to shorten the distance between
mobile device and cloud is to migrate the remote code and
data to the computing resources near to the mobile device
via live migration of the VM from the cloud [134]. However,
live migration of VM is a non-trivial task that requires great
deal of research and development, particularly in networking
environment due to several issues such as large VM size,
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 13
TABLE V
COMPARISON OF CLOUD-BASED SERVERS.
Distant clouds Proximate immobile Proximate mobile Hybrid
computing entities computing entities
Architecture Distributed
Ownership Service provider Public Individual Hybrid
Environment Vendor Premise Business Center Urban Area Hybrid
Availability High Medium Medium High
Scalability High Medium Medium High
Sensing Capabilities Medium Low High High
Utilization Cost Pay-As-You-Use
Computing Heterogeneity High Medium High High
Computing Flexibility High Medium High High
Power Ef?ciency High Medium Medium High
Execution Performance High Medium Medium High
Security and Trust High Moderate Low High
Utilization Rate High
Execution Platform VM VM Physical/VM Physical/VM
Resource Intensity High Moderate Moderate Rich
Complexity Low Moderate Moderate High
Communication Technology 3G/WiFi WiFi WiFi 3G/WiFi
Communication Latency High Low Low Moderate
Execution Latency Low Medium Medium Low
Maintenance Complexity Low Medium Medium High
hard-to-predict user mobility pattern, and limited, intermittent
wireless bandwidth.
Resource utilization is enhanced in clouds due to the virtual-
ization technology deployment. Several VMs can be executed
on a single host to increase the utilization ef?ciency of the
clouds, while each computation task runs on a single isolated
VM loaded on a physical machine. However, VM security
attacks such as VM hopping and VM escape [135] can violate
the code and data security. VM hopping is a virtualization
threat to exploit a VM as a client and attack other VM(s) on
the same host. VM escape is the state of compromising the
security of the hypervisor and control all the VMs.
B. Proximate Immobile Computing Entities
The second type of cloud-based computing resources in-
volves stationary computers located in the public places near
the mobile nodes. The number of computers in public places
such as shopping malls, cinema halls, airports, and coffee
shops is rapidly increasing. These machines are hardly per-
forming tense computational tasks and are mostly playing mu-
sic, showing advertisement, or performing lightweight appli-
cations. Moreover, they are connected to the power socket and
wired Internet. Therefore, it is feasible to leverage such abun-
dant resources in vicinity and perform extensive computation
on behalf of resource-constraint mobile devices. It can also
reduce latency and wireless network traf?c while increases
resource utilization toward green computing. Another group of
proximate immobile computers are Mobile Network Operators
(MNO) and their authorized dealers scattered in urban and
rural areas, private clouds, and public computing kiosk [136]
that can be exploited in smartphone augmentation.
However, protecting security and privacy of mobile user and
computer owner hinder utilization of such nearby resources.
Several shortcomings such as insuf?cient on-premise security
infrastructure, lake of tight security mechanisms, and inef-
?cient update and maintenance procedures inhibit utilizing
such resources (except MNOs) for CMA approaches. Owners
of these resources may attack mobile users and access their
private data on the mobile devices or falsify of?oading results.
Also, malicious users may leverage these resources as an
attacking point to violate mobile users’ security and privacy.
On the other hand, security and privacy of resource owners
are also susceptible to violation. Owners of computer devices
participating in resource sharing require robust mechanisms
to protect and isolate the guest code and data from their
host applications and data. Virtualization aims to realize such
isolation mechanism, but issues such as VM hopping and VM
escape require to be addressed before its successful adoption
[135]. Among all proximate immobile resources, MNOs may
be considered unique in terms of security and privacy features.
MNOs, in general, have been serving mobile users for long
time and could establish high degree of trust among mobile
users. It is feasible to assume that MNO’s certi?ed dealers also
can inherit MNO’s trust if central management and monitoring
process is undertaken by MNOs [46].
C. Proximate Mobile Computing Entities
In this category of cloud-based infrastructures, various
mobile devices, particularly smartphones, Tablets, notebooks,
wearable computers, and handheld computing devices play
the role of servers based on cloud computing principles. The
main bene?t of utilizing nearby mobile resources is their
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 14
Fig. 5. The Hybrid Cloud Concept for MCC.
proximity to the mobile clients. Also, hardware and platform
heterogeneity [51] between mobile servers and clients can be
mitigated, because both sides are mainly ARM-based devices
with mobile OSs. Moreover, contemporary smartphones are
able to provide value added context- and social-aware ser-
vices [137], [138] that contribute to the context-awareness
of mobile applications. However, mobile devices’ resources
are limited and they are unable to perform intensive context-
computing [139]. Realizing distributed computing on cluster
of nearby mobile devices requires several issues, particularly
application architectures, resource scheduling, and mobility to
be addressed.
Moreover, security and privacy of mobile devices as a
service provider is a critical concern in CMA. Mobile devices
are intrinsically susceptible to loss and robbery, and their con-
straint resources inhibit exploiting robust security mechanisms
inside the device. Furthermore, with ever-increasing popularity
of mobile Apps (i.e., mobile applications) in online App stores
such as Google Play
21
and Samsung APPs
22
[140] number
of mobile security threats are rising sharply and malware-
contaminated Apps are becoming serious threats to the mobile
users [141]. Several security threats have been identi?ed in an
experiment of Android mobile applications with the potential
to violate the security of mobile users [142]. Risk of such
contaminated codes can likely be transferred to the non-
contaminated mobile devices by utilizing their computation
resources and request for results of a remote computation.
Hence, establishing trust between mobile devices and end-
users becomes a challenging task.
21https://play.google.com/store
22http://www.samsungapps.com
D. Hybrid (Converged Proximate and Distant Computing En-
tities)
Hybrid infrastructures as depicted in Figure 5 are comprised
of various proximate and distant computing nodes, either
mobile or immobile. The main idea behind building hybrid
resources is to employ heterogeneous computing resources to
create a balance between user requirements (mainly latency
and computation power) and available options [143]. The
latency sensitive codes are of?oaded to the nearest computing
device(s) whereas the most intensive and least latency sensitive
tasks are migrated to the furthest resources. Perhaps, the
utilization costs of nearby resources are more than the remote
servers.
Bene?cial characteristics of hybrid resources summarized
in Table V advocates their usefulness in maximizing the aug-
mentation bene?ts. However, deployment, management, and
resource scheduling processes in dynamic mobile environment
are non-trivial tasks. Developing an autonomic management
system similar to CometCloud [144] in cloud computing and
MAPCloud [143] in MCC to automatically manage, optimize,
and adapt hybrid infrastructures in the cloud-mobile applica-
tions can signi?cantly improve the quality of hybrid CMA
approaches.
Hybrid cloud infrastructures can deliver enhanced security
and privacy features to the CMA approaches and increase the
QoS. Hybrid clouds are comprised of resources with varied
security, privacy, and trust features which can be ef?ciently
utilized by CMA and mobile users as a trade-off. For instance,
security sensitive computations can perform a security-latency
trade-off and execute computation inside a secure distant
cloud.
V. THE STATE-OF-THE-ART CMA APPROACHES:
TAXONOMY
Cloud-based Mobile Augmentation (CMA) is the-state-of-
the-art mobile augmentation model that leverages cloud com-
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 15
Fig. 6. Taxonomy of State-of-the-art CMA Models.
puting technologies and principles to increase, enhance, and
optimize computing capabilities of mobile devices by exe-
cuting resource-intensive mobile application components in
the resource-rich cloud-based resources. According to our
resource classi?cations in previous Section, we analyze and
taxonomize the state-of-the-art CMA approaches into four
models, namely distant ?xed, proximate ?xed, proximate
mobile, and hybrid which are depicted in Figure 6. For
each model, we describe few CMA efforts and tabulate the
comparison results in Figure 10.
A. Distant Fixed
Majority of CMA approaches [25], [27], [29], [31], [33]–
[35], [41], [43], [44], [54], [145] leverage ?xed cloud in-
frastructures in distance due to its straightforward approach.
Utilizing stationary cloud eliminates several management com-
plexities (e.g., resource discovery and scheduling for mobile
cloud-based servers) and alleviates reliability and security
concerns [18]. Works in this class of CMA systems aim
at reducing the complexity and overhead of utilizing distant
cloud. For instance, in [54] authors propose an energy-ef?cient
of?ine job scheduling model based on makespan minimization
model to enhance energy ef?ciency of distant ?xed CMA
systems. Their main notion is to separate the data transmission
from the job execution. During their work, authors provide
several optimization solutions aiming to reduce the energy
consumption of the device during the of?oading process.
However, for the sake of simplicity, the authors study the
energy consumption of tasks in of?ine mode only which does
not consider runtime dynamism of MCC.
Exploiting cloud resources is feasible in several real sce-
narios such as live cloud streaming [98], enterprise appli-
cations (e.g., Customer Relation Management (CRM) and
enterprise resource planning [146]), and Social Networking.
Cloud streaming mechanism has already described in II-C as
an example of utilizing distance ?xed resources. In [146],
researchers leverage cloud resources in developing a CRM
application to enhance ef?ciency of sale representatives for a
pharmaceutical company. The representative meets the physi-
cian in medical centers to promote drugs, present samples and
promotions material, and he records all sale results and details
through the mobile application. The huge database of the
company is stored inside the cloud and the sale representative
can request to process, get, or update data in database without
storing data locally.
We describe some of the distant ?xed CMA approaches that
utilize distant ?xed cloud resources for mobile augmentation
as follows. The terms immobile, ?xed, and stationary are
interchangeably used with the same notion.
• CloneCloud: CloneCloud [34] is a cloud-based, ?ne-
grained, thread-level, application partitioner and execu-
tion runtime that clones entire mobile platform into the
cloud VM and runs the mobile application inside the VM
without performing any change in the application code.
The CloneCloud enables local execution of remaining
mobile application when remote server is running the
intensive components unless local execution tries to ac-
cessing the shared memory state. Cloud resources in this
effort simulate distributed execution of a monolithic ap-
plication in a resourceful environment without engaging
application developer into the distributed application pro-
gramming domain. CloneCloud can signi?cantly reduce
the overall execution time using thread-level migration.
When the local execution reaches the intensive compo-
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 16
nent(s), the CloneCloud system of?oads the component(s)
to the cloud and continues local execution until the
application fetches data from the migrated state. The local
execution is paused until the results are returned and
integrated to the local application.
However, the communication overhead of transferring the
clone of mobile platform, application, and memory state
and frequent synchronization of the shared data between
the mobile and cloud can shrink the power of cloud.
Such overhead becomes more intense in case of heavy,
data- and communication-intensive, and tightly coupled
mobile applications where an alternative execution of
resource-intensive and lightweight components exists.
Frequent code and data encapsulation and migration, and
mobile-cloud data synchronization excessively increase
the communication traf?c and impact on execution time
and energy ef?ciency of the of?oading.
• Elastic Application: Elastic application model [33] is a
CMA proposal leverages distant ?xed cloud data cen-
ter for executing resource-intensive components of the
mobile application. Authors in this model partition a
mobile application into several small components, called
weblet. Weblets are created with least dependency to each
other to increase system robustness while decrease the
communication overhead and latency. The weblet execu-
tion is dynamically con?gured to either perform locally
or remotely, based on the weblet’s resource intensity,
execution environment quality, and of?oading objectives.
The distinctive attribute of this proposal is that application
execution can be distributed among more than one ma-
chine and cooperative results can be pushed back to the
device. To achieve such goal, multiple elasticity patterns
namely replication, splitter, and aggregator are de?ned. In
replication pattern, multiple replicas of a single interface
are executed on multiple machines inside the cloud.
Hence, failure in one replica will not compromise the
system performance. In splitter pattern, the interface and
implementation are separated so that several weblets with
varied implementations can share a single interface. In
aggregator, the results of multiple weblets are aggregated
and pushed to the device for optimized accuracy and
ef?cacy.
The authors endeavor to specify the execution con?gu-
rations (specifying where to run the weblets) at runtime
to match the requirements of the applications and users.
To enhance the overall execution performance and enrich
user experience, the system is able to run the weblets both
locally and remotely. A weblet can be executed remotely
in a low-end device while the same can be executed
locally on a high-end device.
Elastic application model pays more attention to the
user preferences by enabling different running modes of
a single application (e.g., high speed, low cost, of?ine
mode). However, it engages application developers to
determine weblets organization based on the functional-
ity, resource requirements, and data dependency. But, the
characteristics of the weblets are mainly inherited from
the well-known web services to decrease the programmer
burdens.
• Virtual Execution Environment(VEE): Hung et al. [28]
propose a cloud-based execution framework to of?oad
and execute the intensive Android mobile applications
inside the distant cloud’s virtual execution environment.
The quality and accuracy of execution environment is
highly in?uenced by the comprehensiveness and accuracy
of emulated platform. This method uses a software agent
in both mobile and cloud sides to facilitate the overall
system management. The agent in mobile device initiates
VM creation and clones the entire application (even na-
tive codes and UI components) and partial data/memory
state from device to the cloud. Unlike CloneCloud, VEE
aims to reduce latency by migrating the segment of data
stack explicitly created and owned by the application to
the VM instead of copying the entire memory; cloning
the entire memory state, especially for heavy applications
signi?cantly increases latency and traf?c.
During remote execution, the system frequently synchro-
nizes the changes between device and cloud to keep
both copies updated. In order to increase the quality and
ef?ciency of remote execution in virtual environment and
avoid data input loss at application suspension stages,
the system stores input events (reading a ?le, capturing a
face, storing a voice) exploiting a record/replay scheme
and pseudo checkpoint methods. However, these methods
engage application developers to separate the application
state into two states, namely global and local and to
specify the global data structures. The global state con-
tains the program domain and application ?ow, whereas
the local state contains local data structures required by
a method. Programmer usually needs to identify global
state when the application is paused. Once the application
is suspended, the global state will be loaded to avoid re-
execution and the latest Android checkpoint is applied
to the system to re?ect all the changes made from the
last checkpoint. However, all changes, especially user
input might be lost from the last checkpoint. To record
the changes after the last checkpoint, the record/replay
mechanism is deployed by creating a pseudo checkpoint.
To create a pseudo checkpoint, the application noti?es
the local agent to identify the input events and record re-
quired information. Upon the application resumption, the
pseudo checkpoint is restored to restore the application
to the state prior to the suspension.
In this effort, code security inside the cloud is enhanced
by exploiting encryption and isolation approaches that
protects of?oaded code from cloud vendors eavesdrop-
ping. Using probabilistic communication QoS technique,
this is aimed to provide a communication-QoS trade-
off. For instance, the control data (usually small vol-
ume) needs highest accuracy while video streaming data
(often large volume) requires less communication accu-
racy. Moreover, the authors are optimistic that offering
secondary tasks such as automatic virus scanning, data
backup, and ?le sharing in the virtual environment can
enhance quality of user experience.
Although this approach aims to enhance the quality of
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 17
application execution and augment computation capabil-
ity of mobile clients and save energy, but responsiveness
in interactive applications are likely low due to remote UI
execution. Instead of migrating entire application to the
cloud, it might be more bene?cial to utilize some of the
local mobile resources instead of treating mobile device
as a dump client. Data passing between mobile device and
cloud for interactive applications might degrade quality
of experience, especially in low-bandwidth, intermittent
networks.
• Virtualized Screen: Virtualized screen [42] is another
example of CMA approaches that aims to move the screen
rendering process to the cloud and deliver the rendered
screen as an image to the mobile device. The authors
aim to enrich the user experience and migrate the screen
rendering tasks to the cloud with the assumption that
majority of computation- and data-intensive processing
take place in the cloud. Hence, abundant cloud resources’
exploitation simpli?es the CMA system architecture,
prolongs mobile battery, and enhances the interaction
and responsiveness of mobile applications toward rich
user experience. Screen virtualization technique (running
partial rendering in cloud and rest in mobile depending
on the execution context) is envisioned to optimize user
experience, especially for lightweight, high-?delity, in-
teractive mobile applications that entirely run on local
resources. Their conceptual proposal aims to enhance
visualization capability of mobile clients, mitigate the
impact of hardware and platform heterogeneity, and facil-
itate porting mobile applications to various devices (e.g.,
smartphone, laptop, and IP TV) with different screens.
To reduce the mobile-cloud data transmission, a frame-
based representation system is exploited to forward the
screen updates from the cloud to the mobile. Frame-based
representation system captures and feeds the whole screen
image to the transmission unit. This approach updates
each frame based on the previous frame stored inside
both the mobile and cloud. However, a rich interactive,
responsive GUI needs live streaming of screen images
which is impacted by communication latency. Although
the authors describe optimized screen transmission ap-
proaches to reduce the traf?c, the impact of computation
and communication latency is not yet clear, as this is
a preliminary proposal. Moreover, utilizing virtualized
screen method for developing lightweight mobile-cloud
application is a non-trivial task in the absence of its
programming API.
• Cloud-Mobile Hybrid (CMH) Application: Unlike appli-
cation of?oading solutions, authors in this proposal [32]
introduce a new approach of utilizing cloud resources
for mobile users. In this effort, the authors propose a
novel CMH application model, in which heavy compo-
nents are developed for cloud-side execution, whereas
lightweight or native codes are developed for mobile
devices execution. CMH Applications execution does not
need pro?ling, partitioning, and of?oading processes and
hence produce least computation overhead on mobile
devices. Upon successful cloud-side execution, the results
are returned back to the mobile for integrating to the
native mobile components.
However, developing CMH applications is signi?cantly
complex due to the interoperability and vendor lock-in
problems in clouds and fragmentation issue in mobiles
[51]. Cloud components designed for a speci?c cloud are
not able to move to another cloud due to underlying het-
erogeneity among clouds. Similarly, mobile components
developed for a particular platform cannot be ported to
different platforms because of heterogeneity. Yet isolating
development of mobile and cloud components creates
further versioning and integration challenges.
To mitigate the complexity of CMH application devel-
opments and facilitate portability, the authors leverage
Domain Speci?c Language (DSL) [147], [148]. A DSL
is a programming language with major focus on solving
problem in speci?c domains. MATLAB
23
is a well-known
DSL-based tool for mathematicians. A parser takes a DSL
script and converts codes into an in-memory object to be
forwarded to various automatic component generators.
The system needs different code generators for various
mobile and cloud platforms. Once the mobile and cloud
components are generated, the CMH application can be
assembled for various mobile-cloud platforms. However,
utilizing DSL-based techniques requires more generaliza-
tion efforts to be bene?cial in developing all types of
CMH applications.
• µCloud: Similar to the CMH framework, µCloud [36] is
a modular, mobile-cloud application framework that aims
to facilitate mobile-cloud application generation, promote
application portability, minimize the development com-
plexity, and enhance of?ine usability in intensive mobile-
cloud applications. Ful?lling separation of concerns vi-
sion, skilled programmers independently develop self-
contained components which do not have any direct inter
communications with each other. Unskilled mobile users
can mash-up (assembling available components to build
complex application) these prefabricated components to
generate a complex mobile-cloud application. Cloud ven-
dors provide infrastructure and platform as cloud services
to run prefabricated cloud components. The main idea in
this proposal is to avoid local execution of the resource-
intensive components. Hence, components are identi?ed
as cloud, mobile, and hybrid; mobile components are
executable exclusively on mobile and cloud components
are strictly developed for cloud server while hybrid com-
ponents can either run locally or remotely. Hybrid com-
ponents have either multiple implementations or a single
implementation that need a middleware for execution.
Each component has a triplet of identi?er, input/output
parameters, and con?guration.
To alleviate of?ine usability issue, the authors leverage
mobile-side queuing and cloud-side caching to main-
tain data in case of disconnection. Data will be trans-
ferred upon reconnection. Application is partitioned into
components and organized as a directed graph. Nodes
23http://www.mathworks.com/products/matlab/
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 18
represent components and vertices indicate data/control
?ow. Application is divided into three fragments; in each
fragment, a managing unit called orchestrator executes
and maintains component’s mash-up process. The output
of each component is forwarded using the pass-by-value
semantic as an input to the subsequent component.
Unlike Elastic Application model, the design and im-
plementation of components in µCloud is statically per-
formed in early development phase. Thus, any improve-
ment in resource availability of mobile devices or envi-
ronmental enhancement (like bandwidth growth) will not
improve the overall execution of µCloud applications.
Such in?exibility decreases the application execution
performance and degrades the quality of user experience.
• SmartBox: Smartbox [112] is a self-management, on-
line, cloud-based storage and access management system
developed for mobile devices to expand device storage
and facilitate data access, and sharing. It is a write-
once, read many times system designed to store personal
data such as text, song, video, and movies which is
not appropriate for large scale computational datasets. In
Smartbox, mobile devices are associated with a shadow
storage to store/retrieve personal data using a unique
account. To facilitate data sharing among larger group
of end-users in of?ce or at home, a public storage space
is provisioned.
Smartbox exploits traditional hierarchical namespace
for smooth navigation and employs an attribute-based
method to facilitate data navigation and service query
using semantic metadata such as the publisher-provider
metadata. Data navigation and query using tiny keyboard
and small screen irk mobile users when inquiring and
navigating stored data in cloud. However, mobile users
need always-on connectivity to access online cloud data
which is not yet achieved and is unlikely to become
reality in near future.
• WhereStore: WhereStore [149] is a location-based data
store for cloud-interacting mobile devices to replicate
necessary cloud-stored mobile data on the phone. The
main notion in this effort is that users in different places
doing various activities need dissimilar types of informa-
tion. For instance, a foreign tourist in Manhattan requires
information about nearby places of interest rather than all
the country. Hence, identifying the location and caching
predicted data deemed can enhance the system ef?ciency
and user experience. However, ef?cient prediction of
future user location and required data, and determining
the right time for caching data are challenging tasks.
• Wukong: Wukong [150] is a cloud oriented ?le service
for multiple mobile devices as a user-friendly and highly
available ?le service. The authors provision support of
heterogeneous back-end services such as FTP, Mail, and
Google Docs Service in a transparent manner leveraging a
service abstraction layer (SAL). Wukong enables appli-
cations to access cloud data without being downloaded
into the local storage of mobile device.
Authors introduce cache management and pre-fetch
mechanisms in different scenarios to increase perfor-
mance while decreasing latency. However, it cannot al-
ways reduce latency due to the bandwidth limitation and
I/O overhead. In operations with long gap between open
and read, it is bene?cial to pre-fetch data from cloud
to the device that signi?cantly improves user experience.
Data security is enhanced via an encryption module
and bandwidth is saved using a compression module.
While compression methods utilized in this proposal is
inef?cient for multimedia ?les like image and music, it
can compress text and log ?les noticeably.
We conclude that one of the most effective solutions to
tackle bandwidth and latency limitations in CMA ap-
proaches, especially cloud storage is to decrease the vol-
ume of data using imminent compression methods. While
various compression methods work well on speci?c ?le
types, a cognitive or adaptive compression method with
focus on multimedia ?les can signi?cantly improve the
feasibility of cloud-storage systems.
B. Proximate Fixed
Researchers have recently proposed CMA approaches in
which nearby stationary computers are utilized. Utilizing
nearby desktop computers initiates new generation of services
to the end-user via mobile device. In [26], the authors pro-
vide a real scenario in which Ron, a patient diagnosed with
Alzheimer, receives cognitive assistance using an augmented-
reality enabled wearable computer. The system consists of a
lightweight wearable computer and a head-up display such
as Google Glass
24
equipped with a camera to capture the
environment and an earphone to send the feedback to the
patient. The system captures the scene and sends the image to
the nearby ?x computers to interpret the scene in the image
using the object or face recognition, voice synthesizer, and
context-awareness algorithms. When Ron looks at a person for
few seconds, the person’s name and some clue information
is whispered in Ron’s ear to help greeting with the person.
When he looks at his thirsty plant or hungry dog, the system
reminds Ron to irrigate the plant and feed his dog. The nearby
resources are core component of this system to provide low-
latency real-time processing to the patient. In this part, we
explain one of the most prominent proximate ?xed efforts as
follows.
• Cloudlet: Cloudlet [26] is a proximate immobile cloud
consists of one or several resource-rich, multi-core, Gi-
gabit Ethernet connected computer aiming to augment
neighboring mobile devices while minimizing security
risks, of?oading distance (one-hop migration from mo-
bile to Cloudlet), and communication latency. Mobile
device plays the role of a thin client while the intensive
computation is entirely migrated via Wi-Fi to the nearby
Cloudlet. Although Cloudlet utilizes proximate resources,
the distant ?xed cloud infrastructures are also accessible
in case of Cloudlet scarcity. The authors employ a decen-
tralized, self-managed, widely-spread infrastructure built
on hardware VM technology. Cloudlet is a VM-based
24http://www.google.com/glass/start/
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 19
Fig. 7. Cloudlet-based Resource-Rich Mobile Computing Life Cycle.
of?oading system that can signi?cantly shrink the impact
of hardware and OS heterogeneity between mobile and
Cloudlet infrastructures.
To reduce the Cloudlet management and maintenance
costs while increasing security and privacy of both
Cloudlet host and mobile guest, a method called “tran-
sient Cloudlet customization” is deployed which uses
hardware VM technology. It enables Cloudlet customiza-
tion prior to the of?oading and performs Cloudlet restora-
tion as a post-of?oading cleanup process to restore the
host to its original software stake. The VM encapsulates
the entire of?oaded mobile environment (data state and
code) and separates it from the host permanent software.
Hence, feasibility of deploying Cloudlet in public places
such as coffee shops, airport lounge, and shopping malls
increases.
Unlike CloneCloud and Virtual Execution Environment
efforts that migrate the entire mobile OS clone to the
cloud, Cloudlet assumes that the entire OS clone exists
and is preloaded in the host and runs on an isolated
VM. In mobile side, instead of creating the VM of
the entire mobile application and its memory stack, the
systems encapsulates a lightweight software interface of
the intensive components called VM overlay.
The VM overall of?oading performance is further en-
hanced by exploiting Dynamic VM Synthesis (DVMS)
method since its performance solely depends on
the mobile-Cloudlet bandwidth and cloudlet resources.
DVMS assumes that the base VM is already available
in the target Cloudlet and user can ?nd the match-
ing execution environment (VM base) among silo of
nearby Cloudlets. Upon discovery and negotiation of the
Cloudlet, the DVMS of?oads the VM overlay to the
infrastructure to execute launch VM (base + overlay).
Henceforth, the of?oaded code starts execution in the
state it was paused. Upon completion of Cloudlet execu-
tion the VM residue is created and sent back to the mobile
device. In the Cloudlet, the VM is discarded as a post-
of?oading cleanup process to restore the original Cloudlet
state. In mobile side, the results will be integrated to
the application and local execution will be resumed. To
present a clear understanding of the overall process, the
sequence diagram of Cloudlet-based resource-rich mobile
computing is depicted in Figure 7.
Despite the noticeable of?oading improvements in the
Cloudlet, its success highly depends on the existence of
plethora of powerful Cloudlets containing popular mobile
platforms’ base VM. Encouraging individual owners to
deploy such Cloudlets in the absence of monetary incen-
tives is an issue that must be addressed before deployment
in real scenarios. Although energy ef?ciency, security
and privacy, and maintenance of Cloudlet are widely
acceptable, further efforts are required to protect the
overall CMA process. Moreover, few minutes of?oading
latency in Cloudlet is unacceptable to users [151].
C. Proximate Mobile
Recently, several researchers [24], [45], [122], [152]–[155]
propose CMA approaches in which nearby mobile devices
lend available resources to other mobile clients for execution
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 20
Fig. 8. MOMCC Concept.
of resource-intensive tasks in distributed manner. Utilizing
such resources can enhance user experience in several real
scenarios such as Optical Character Recognition (OCR) and
natural language processing applications. The feasibility of
utilizing nearby mobile devices is studied in [24] where Peter,
a foreign tourist, visiting a Korean exhibition and ?nds interest
in an exhibit, but cannot understand the Korean description.
He can take a photo of the manuscript and translate it using
the OCR application, but his device lacks enough computation
resources. Although he can exploit the Internet web services
to translate the text, the roaming cost is not affordable to him.
Hence, he leverages a CMA solution by utilizing computation
resources of nearby mobile devices to complete the task. Some
of the CMA efforts whose remote resources are proximate
mobile devices are explained as follows.
• MOMCC: Market-Oriented Mobile Cloud Computing
(MOMCC) [45] is a mobile-cloud application frame-
work based on Service Oriented Architecture (SOA)
that harnesses a cluster of nearby mobile devices to
run resource-intensive tasks. In MOMCC, mobile-cloud
applications are developed using prefabricated building
blocks called services developed by expert programmers.
Service developers can independently develop various
computation services and uploaded them to a publicly
accessible UDDI (Universal Description Discovery and
Integration) such as mobile network operators.
Services are mostly executed on large number of smart-
phones in vicinity which can share their computation
resources and earn some money. To enhance resource
availability and elasticity, distant stationary cloud re-
sources are also available if nearby resources are in-
suf?cient. In order to become an IaaS (Infrastructure
as a Service) provider, mobile devices register with the
UDDI and negotiate to host certain services after secure
authentication and authorization. Mobile users at runtime
contact UDDI to ?nd appropriate secure host in vicinity
to execute desired service on payment. The collected rev-
enue is shared between service programmer, application
developer, UDDI, and service host for promotion and
encouragement. Figure 8 depicts the MOMCC concept.
However, MOMCC is a preliminary study and its overall
performance is not yet evident. Several issues are required
to be addressed prior to its successful deployment in
real scenarios. Executing services on mobile devices is a
challenging task considering resource limitation, security,
and mobility. Also an ef?cient business plan that can
satisfy all engaging parties in MOMCC is lacking and
demand future efforts. MOMCC can provide an income
source for mobile owners who spend couple of hundred
dollars to buy a high-end device. In addition, faulty
resource-rich mobile devices that are able to function
accurately can be utilized in MOMCC instead of being
e-waste.
• Hyrax: Hyrax [152] is another CMA approach that ex-
ploits the resources from a cluster of immobile smart-
phones in vicinity to perform intense computations.
Hyrax alleviates the frequent disconnections of mobile
servers using fault tolerance mechanism of Hadoop. Sim-
ilar to MOMCC and Cloudlet, due to resource limita-
tions of smartphone servers, the accessibility to distant
stationary clouds is also provisioned in case the nearby
smartphone resources are not suf?cient. However, Hyrax
does not consider mobility of mobile clients and mobile
servers. Hence, deployment of Hyrax in real scenarios
may become less realistic due to immobilization of mo-
bile nodes. Lack of incentive for mobile servers also
hinders Hyrax success.
Hyrax is a MCC platform developed based on Hadoop
[156] for Android smartphones. In developing Hyrax,
the MapReduce [157] principles are applied utilizing
Hadoop as an open source implementation of MapRe-
duce. MapReduce is a scalable, fault-tolerant program-
ming model developed to process huge dataset over a
cluster of resources. Centralized server in Hyrax runs two
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 21
client side processes of MapReduce, namely NameNode
and JobTracker processes to coordinate the overall com-
putation process on a cluster of smartphones. In smart-
phone side, two Hadoop processes, namely DataNode
and TaskTracker are implemented as Android services
to receive computation tasks from the JobTracker. Smart-
phones are able to communicate with the server and other
smartphones via 802.11g technology.
Nevertheless, the cloud storage connectivity in Hyrax is
missing. It demands several gigabytes of local storage to
store data and computation. Hence, user cannot access
distributed data over the Internet or Ethernet. The author
utilizes the constant historical multimedia data to avoid
?le sharing. Hence, it is less bene?cial for interactive
and event-oriented applications whose data frequently
changes over the execution and also data-intensive appli-
cations that require huge database. The overall overhead
in Hyrax is high due to the intensity of Hadoop algorithm
which runs locally on smartphones.
• Virtual Mobile Cloud Computing (VMCC): Researchers
in [24] aim to augment computing capabilities of stable
mobile devices by leveraging an ad-hoc cluster of nearby
smartphones to perform intensive computing with min-
imum latency and network traf?c while decreasing the
impact of hardware and platform heterogeneity. During
the ?rst execution, required components (proxy creation
and RPC support) are added to the application code to
be used for of?oading; the modi?ed code will remain for
future of?oading. For every application, the system de-
termines the number of required mobile servers, security
and privacy requirements, and of?oading overhead. The
system continuously traces the number of total mobile
servers and their geolocation to establish a peer-to-peer
communication among them. Upon decision making the
application is partitioned into small codes and transferred
to the nearby mobile nodes for execution. The results will
be reintegrated back upon completion.
However, several issues encumber VMCC’s success.
Firstly, this solution, similar to Hyrax, is not suitable
for a moving smartphone since the authors explicitly
disregard mobility trait of mobile clients. Secondly, every
computing job is sent to exactly one mobile node; so, the
of?oading time and overhead will be increased when the
serving node leaves the cluster. Thirdly, the of?oading
initiation might take long since the of?oading’s overall
performance highly depends on the number of available
nearby nodes; insuf?cient number of mobile nodes defers
of?oading. Finally, in the absence of monetary incentive
for mobile nodes the likelihood of resource sharing
among resource-constraint mobile devices is low.
D. Hybrid
Hybrid CMA efforts are budding [46], [143], [158] to opti-
mize the overall augmentation performance and researchers are
endeavoring to seamlessly integrate various types of resources
to deliver a smooth computing experience to mobile end-
users. For instance, mCloud [159] is an imminent proposal to
integrate proximate immobile and distant stationary computing
resources. Authors are aiming to enable mobile-users to per-
form resource-intensive computation using hybrid resources
(integrated cloudlet-cloud infrastructures). Hybrid solutions
aim to provide higher QoS and richer interaction experience
to the mobile end-users of real scenarios explained in previous
parts. For instance, in the foreign tourist example, the image
can be sent to the nearby mobile device of a non-native local
resident for processing. When the processing fails due to lack
of enough resources, the picture can be forwarded to the cloud
without Peter pays high cost of international roaming (Peter
may pay local charge).
We review some of the available hybrid CMAs as follows.
• SAMI: SAMI (Service-based Arbitrated Multi-tier In-
frastructure for mobile cloud computing) [46] proposes
a multi-tier IaaS to execute resource-intensive compu-
tations and store heavy data on behalf of resource-
constraint smartphones. The hybrid cloud-based infras-
tructures of SAMI are combination of distant immobile
clouds, nearby Mobile Network Operators (MNOs), and
cluster of very close MNOs authorized dealers depicted
in Figure 9. The compound three level infrastructures aim
to increase the outsourcing ?exibility, augmentation per-
formance, and energy ef?ciency. The MNO’s revenue is
hiked in this proposal and energy dissipation is prevented.
Nearby dealers can be reached by Wi-Fi. MNO’s can be
accessed either directly via cellular connection or through
dealers via Wi-Fi and broadband. Connection is estab-
lished via cellular network to contact distant stationary
clouds. The cluster of nearby stationary machines (MNO
dealers located in vicinity) performs latency-sensitive
services and omits the impact of network heterogeneity.
SAMI leverages Wi-Fi technology to conserve mobile
energy because it consumes less energy compared to
the cellular networks [116]. In case of nearby resource
scarcity or end-user mobility, the service can be executed
inside the MNO via cellular network. However, if the
resources in MNO are insuf?cient, the computation can
be performed inside the distant immobile cloud.
The resource allocation to the services is undertaken by
arbitrator entity based on several metrics, particularly
resource requirements, latency, and security requirements
of varied services. The arbitrator frequently checks and
updates the service allocation decision to ensure high
performance and avoids mismatch.
To enhance security of infrastructures, SAMI employs
comparatively reliable and trustworthy entities, namely
clouds, MNOs, and MNO trusted dealers. MNOs have
already established reputation-trust among mobile users
and can enforce a strict security provisions to establish
indirect trust between dealers and end users ensuring that
user’s security and privacy will not be violated. SAMI ap-
plication development framework facilitates deployment
of service-based platform-neutral mobile applications and
eases data interoperability in MCC due to utilization of
SOA.
However, SAMI is a conceptual framework and deploy-
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 22
Fig. 9. SAMI: A Multi-Tier Cloud-based Infrastructure Model.
ment results are expected to advocate its feasibility. SAMI
imposes a processing overhead on MNOs due to con-
tinuous arbitration process. Deployment, management,
and maintenance costs of SAMI are also high due to
the existence of various infrastructure layers. Moreover,
though the authors discuss the monetary aspects of the
proposal, a detailed discussion of the business plan is
missing, for example in what scenario resource outsourc-
ing is affordable for the mobile application? How does
income should be divided among different entities to be
satisfactory?
• MOCHA: In MOCHA [158] authors propose a mobile-
cloudlet-cloud architecture for face recognition applica-
tion using mobile camera and hybrid infrastructures of
nearby Cloudlet and distant immobile cloud. Cloudlet
is a speci?c, cheap cluster of computing entities like
GPU (Graphics Processing Unit) capable of massively
processing data and transactions in parallel. Cloudlets
are able to be accessed via heterogeneous communication
technologies such as Wi-Fi, Bluetooth, and cellular. The
mobile often access processing resources via Cloudlet
rather than directly connecting to the cloud, unless ac-
cessing cloud resources bears lower latency.
Cloudlet receives the smartphones intensive computation
tasks and partitions them for distribution between it-
self and distant immobile clouds to enhance QoS [26].
MOCHA leverages two partitioning algorithms: ?xed
and greedy. In the ?xed algorithm, the task is equally
partitioned and distributed among all available computing
devices (including Cloudlet and cloud servers), whereas
in greedy algorithm, the task is partitioned and distributed
among computing devices based on their response times;
the ?rst partition is sent to the quickest device while the
last partition is sent to the slowest device. The response
time of the task partitioned using greedy approach is
signi?cantly better than ?xed, especially when Cloudlet
server is utilized in augmentation process and large
number of clouds with heterogeneous response time exist.
However, smartphones in MOCHA require prior knowl-
edge of the communication and computation latency of
all available computing entities (Cloudlet and all available
distant ?xed clouds) which is a resource-hungry and time-
consuming task.
VI. CMA PROSPECTIVES
People dependency to mobile devices is rapidly increasing
[89], [160] and smartphones have been using in several crucial
areas, particularly healthcare (tele-surgery), emergency and
disaster recovery (remote monitoring and sensing), and crowd
management to bene?t mankind [161]–[163]. However, intrin-
sic mobile resources and current augmentation approaches are
not matching with the current computing needs of mobile-
users, and hence, inhibit smartphone’s adoption. Upon slow
progress of hardware augmentation, the highly feasible solu-
tion to ful?ll people computing needs is to leverage CMA
concept. This Section aims to present set of guidelines for
ef?ciency, adaptability, and performance of forthcoming CMA
solutions. We identify and explain the vital decision making
factors that signi?cantly enhance quality and adaptability of
future CMA solutions and describe ?ve major performance
limitation factors. We illustrate an exemplary decision making
?owchart of next generation CMA approaches.
A. CMA Decision Making Factors
These factors can be used to decide whether to perform
CMA or not and are needed at design and implementation
phases of next generation CMA approaches. We categorize
the factors into ?ve main groups of mobile devices, contents,
augmentation environment, user preferences and requirements,
and cloud servers, which are depicted in Figure 11 and
explained as follows.
1) Mobile Devices: From the client perspective, amount
of native resources including CPU, memory, and storage is
the most important factor to perform augmentation. Also,
energy is considered a critical resource in the absence of long-
spanning batteries. The trade-off between energy consumed by
augmentation and energy squandered by communication is a
vital proportion in CMA approaches [73]. Device mobility and
communication ability (supporting varied technologies such
as 2G,3G,Wi-Fi) are other metrics that are important in the
of?oading performance.
2) Contents: Another in?uential factor for CMA decision
making is the contents’ nature. The code granularity and
size as well as data type and volume are example attributes
of contents that highly impact on the overall augmentation
process. Hence, the augmentation should be performed con-
sidering the nature and complexity of application and data.
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 23
Fig. 10. Comparison of CMA Approaches.
For instance, latency sensitive small data are ef?cient to be
processed locally, whereas sensitive big data are encouraged to
be stored in a large reliable cloud storage. Similarly, of?oading
a coarse-grained, large code to a distant ?xed cloud via a low
bandwidth network is not feasible.
3) Augmentation Environment: Mobile computing is a het-
erogeneous environment comprised of non-uniform mobile
nodes, communication technologies, and resources. One of the
most in?uential environment-dependent factors is the wireless
communication medium in which majority of communications
take place. Wireless is an intermittent, unreliable, risky, and
blipping medium with signi?cant impact on the quality of
augmentation solutions. The overall performance of a low cost,
highly available, and scalable CMA approach is magni?cently
shrunk by the low quality of communication medium and
technologies. Selecting the most suitable technology consider-
ing the factors like required bandwidth, congestion, utilization
costs, and latency [164] is a challenge that affects quality
of augmentation approaches in wireless domains. Wireless
medium characteristics impose restrictions when specifying
remote servers at design time and runtime.
Moreover, dynamism and rapidly changing attributes of the
runtime environment noticeably impact on augmentation pro-
cess and increase decision making complexity. Augmentation
approaches should be agile in dynamic mobile environment
and instantaneously re?ect to any change. For example, user
movement from high bandwidth to a low bandwidth network,
receding from the network access point, and rapidly changing
available computing resources complicate CMA process.
4) User Preferences and Requirements: End-users’ physi-
cal and mental situations, individual and corporate preferences,
and ultimate computing goals are important factors that affect
of?oading performance. Some users are not interested to
utilize the risky channel of Internet, while others may demand
accessing cloud services through the Internet. Hence, users
should be able to modify technical and non-technical spec-
i?cations of the CMA system and customize it according to
their needs. For example, user should be able to alter degree of
acceptable latency against energy ef?ciency of an application
execution. Selecting the most appropriate resource among
available options can also enhance overall user experience.
5) Cloud Servers: As explained, CMA approaches can
leverage various types of cloud resources to enhance com-
puting capabilities of mobile devices. Therefore, the overall
performance and credibility of the augmentation approaches
highly depend on the cloud-based resources’ characteristics.
Performance, availability, elasticity, vulnerability to security
attacks, reliability (delivering accurate services based on
agreed terms and conditions), cost, and distance are major
characteristics of the cloud service providers used for aug-
menting mobile devices.
Utilizing clouds to augment mobile devices notably reduces
the device ownership cost by borrowing computing resources
based on pay-as-you-use principle. Such elastic, cost-effective,
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 24
Fig. 11. Critical Factors in CMA Decision Making.
reliable, and relatively trustworthy resources are embraced by
the scholars, industrial organizations, and end-users towards
?ourishing CMA approaches.
B. Performance Limitation Factors
Performance of varied CMA solutions is impacted by sev-
eral factors. We describe ?x major performance limitation
factors as follows.
1) Heterogeneity: MCC is a highly heterogeneous envi-
ronment comprised of three diversi?ed domains of mobile
computing, cloud computing, and networking. Although het-
erogeneity can provide ?exibility to the mobile users by
providing selection alternatives, it breeds several limitations
and challenges, especially for developing multi-tier CMA-
based applications [51]. Dissimilar mobile platforms such as
Android, iOS, Symbian, and RIM beside diverse hardware
characteristics of mobile device inhibit data and application
portability among varied mobile devices. Portability is the abil-
ity to migrate code and data from one device to another with
no/less modi?cation and change [165]. Existing heterogene-
ity in cloud computing including hardware, platform, cloud
service policy, and service heterogeneity originates challenges
such as portability and interoperability and fragment the MCC
domain.
Network heterogeneity in MCC is the composition of var-
ious wireless technologies such as Wi-Fi, 3G, and WiMAX.
Mobility among varied network environments intensi?es com-
munication de?ciencies and stems complex issues like signal
handover [125]. Inappropriate decision making during the
handover process like (i) less appropriate selection of network
technology among available candidates and (ii) transferring the
communication link at the wrong time, increases WAN latency
and jitter that degrade quality of mobile cloud services. Con-
sequently delay-sensitive content and services are degraded
[166] and adoption of CMA approaches are hindered.
2) Data Volume: Ever-increasing volume of digital con-
tents [85] signi?cantly impacts on the performance of CMA
approaches in MCC. Current wireless infrastructures and tech-
nologies fail to ef?ciently ful?ll the networking requirements
of CMA approaches. Storing such a huge data in a single ware-
house is often impossible and demands data partitioning and
distributed storage that not only mitigates data integrity and
consistency, but also makes data management a pivotal need
in MCC [167]. Applying a single access control mechanism
for relevant data in various storage environments is another
challenging task that impacts on the performance and adoption
of CMA solutions in MCC.
3) Round-Trip Latency: Communication and computation
latency is one of the most important performance metrics of
mobile augmentation approaches, especially when exploiting
distant cloud resources. In cellular communications, distance
from the base station (near or far) and variations in bandwidth
and speed of various wireless technologies affect the perfor-
mance of augmentation process for mobile devices. Moreover,
leveraging wireless Internet networks to of?oad content to the
distant cloud resources creates a bottleneck. Latency adversely
impacts on the energy ef?ciency [73] and interactive response
[168] of CMA-based mobile applications due to excessive
consumption of mobile resources and raising transmission
delays.
Recently, researches [169], [170] are emerging toward de-
creasing the networking overhead and facilitating mobility
(both node and code mobility) in cloud-based of?oading
approaches. For example, Follow-Me Cloud [169] aims at
enabling mobility of network end-points across different IP
IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 25
subnets. The authors employ the concept of identi?er and
locator separation of edge networks using OpenFlow-enabled
switches. Leveraging the Follow-Me Cloud, mobile nodes
can move among access networks without being noti?ed of
any change or session disruption. All corresponding nodes
that have been communicating with the mobile node can
continue their communication without interruption. When the
node migrates, its old IP turn to identi?er and its new IP
address becomes locator address so that all other nodes can
keep communication with the moving node. However, for
each packet traveling to/from the mobile node, there is an
overhead of manipulating the locator/identi?er values. Future
improvement and optimization efforts will enhance the CMA
systems’ performance.
In cloud side, computation latency signi?cantly impacts on
the application responsiveness. Researchers study the impact
of cloud computation performance on the execution time
and vindicate 12X reduction in performance time violation
[171]. Thus, the increased latency degrades the quality of
user experience and adversely impacts on the user-perceived
performance of CMA solutions.
4) Context Management and Processing: Performance of
CMA approaches is noticeably degraded by lack of suf?cient,
accurate knowledge about the runtime environment. Contem-
porary mobile devices are capable of gathering extensive con-
text and social information such as available remote resources,
network bandwidth, weather conditions, and users’ voice and
gestures from their surrounding environment [137], [138]. But,
storing, managing, and processing large volume of context
information (considering MCC environment’s dynamism and
mobile devices’ mobility) on resource-constraint smartphones
are non-trivial tasks.
5) Service Execution and Delivery: SLA as a formal
contract between service consumer and provider enforces
resource-level QoS (e.g., memory capacity, compute unit, and
storage) against a fee, which is not suf?cient for mobile users
in highly dynamic wireless MCC environment. User-perceived
performance in MCC is highly affected by the quality of cloud
computations, wireless communications, and local execution.
Hence, varied service providers, including cloud vendors,
wireless network providers, and mobile hardware and OS
vendors need to collaborate and ensure acceptable level of
QoS. For successful CMA approaches, comprehensive real-
time monitoring process is expected to ensure that engaging
service vendors are delivering required services in acceptable
level based on the accepted SLA.
C. CMA Feasibility
Although CMA is bene?cial and can saves resources [40],
several questions need to be addressed before CMA can be
implemented in real scenarios. For instance: is CMA always
feasible and bene?cial? Can CMA save local resources and
enhance user experience? What kind of cloud-based resources
should be opted to achieve the superior performance?
Vision of future CMA proposals will be realized by accurate
sensing and acquiring precise knowledge of decision making
factors like user preferences and requirements, augmentation
environment, and mobile devices, which are explained in
previous part. A decision making system, similar to those used
in [25], [33], [49], analyzes these vital factors to determine
the augmentation feasibility and speci?es if augmentation can
ful?ll mobile computation requirements and enrich quality
of user experience. Figure 12 illustrates a possible decision
making ?ow of future CMA approaches.
Availability of mobile resources to manage augmentation
process and volume of cloud resources to provision required
resources signi?cantly impact on the quality of augmentation
[9]. Similarly, user preferences, limitations, and requirements
affect the augmentation decision making. For instance, if aug-
mentation is not permitted by users, the application execution
and data storage should be performed locally without being
of?oaded to a remote server(s) or be terminated in the absence
of enough local resources. Similarly, augmentation process
can be terminated if the execution latency of delay-sensitive
content is sharply increased, quality of execution is noticeably
decreased, or security and privacy of users is violated [19].
Furthermore, usefulness of CMA approaches highly de-
pends on the execution environment. Of?oading computation
and mobile-cloud communication ratio, distance from mobile
to the cloud, network technologies and coverage, available
bandwidth, traf?c congestion, deployment cost, and even na-
ture of augmentation tasks alter usefulness of the CMA ap-
proaches [40]. For instance, performing an of?oading method
on a data-intensive application (e.g., applying a graphical ?lter
on large number of high quality images) in a low-bandwidth
network imposes large latency and signi?cantly degrades user
experience which should be avoided. Similarly, migrating a
resource-hungry code to an expensive remote resource can
be unaffordable practice. Suppose in a sample augmentation
approach R
C
is the total native resources consumed during
augmentation, R
M
is the total native resources consumed
for maintenance, and R
S
is the total resources conserved in
augmentation process. Explicitly for a feasible augmentation
approach R
C
+ R
M