Computer Networks 108 (2016) 260–278

Contents lists available at ScienceDirect

Computer Networks

journal homepage: www.elsevier.com/locate/comnet

Cyber-physical systems clouds: A survey

Rihab Chaâri a , b , ∗, Fatma Ellouze

b , c , Anis Koubâa

b , d , f , Basit Qureshi d , Nuno Pereira

f , Habib Youssef e , Eduardo Tovar f

a University of Mannouba, National School of Computer Science and Engineering (ENSI), Tunisia b Cooperative Intelligent Networked Systems (COINS) Research Group, Riadh, Saudi Arabia c University of Sfax, ReDCAD Laboratory, B.P. 1173, Sfax, Tunisia d Prince Sultan University, Saudi Arabia e University of Sousse, PRINCE Research Unit, Sousse, Tunisia f CISTER/INESC-TEC and ISEP, Polytechnic Institute of Porto, Portugal

a r t i c l e i n f o

Article history:

Received 9 September 2015

Revised 12 August 2016

Accepted 17 August 2016

Available online 1 September 2016

Keywords:

Cloud computing

Cloud robotics

Cloud sensors

Vehicular cloud networks

a b s t r a c t

Cyber-Physical Systems (CPSs) represent systems where computations are tightly coupled with the phys-

ical world, meaning that physical data is the core component that drives computation. Industrial au-

tomation systems, wireless sensor networks, mobile robots and vehicular networks are just a sample of

cyber-physical systems. Typically, CPSs have limited computation and storage capabilities due to their tiny

size and being embedded into larger systems. With the emergence of cloud computing and the Internet-

of-Things (IoT), there are several new opportunities for these CPSs to extend their capabilities by taking

advantage of the cloud resources in different ways. In this survey paper, we present an overview of re-

search effort s on the integration of cyber-physical systems with cloud computing and categorize them

into three areas: (1) remote brain, (2) big data manipulation, (3) and virtualization. In particular, we fo-

cus on three major CPSs namely mobile robots, wireless sensor networks and vehicular networks.

© 2016 Elsevier B.V. All rights reserved.

c

s

r

s

e

a

p

(

m

s

e

P

i

h

[

c

a

m

1. Introduction

Internet-of-Things (IoT) and Cloud computing are nowadays the

two most prominent ICT technologies that are integrated in differ-

ent applications areas. The IoT paradigm represents the new In-

ternet generation where cyber-physical devices such as wireless

sensor nodes, RFID-based devices, and mobile robots are intercon-

nected together through the web. This integration extends the role

of the Internet from carrying out pure digital data to handling

physical data emanating from these cyber-physical devices. While

the IoT brings a lot of new pervasive applications, there are mainly

two major challenges that must be addressed: (1) the process-

ing of the large amount of data emerging from these devices, in

particular considering their limited computation and storage capa-

bilities, (2) and the seamless access to these devices by Internet

users through different channels. Nowadays, the cloud computing

paradigm represents a primary solution to these challenges in tra-

ditional distributed computing systems as it seamlessly provides

abundant resources for client applications in terms of computation

and storage resources, with the accounting model “pay as you go”.

∗ Corresponding author.

E-mail address: rihab.chaari@coins-lab.org (R. Chaâri).

b

T

r

t

http://dx.doi.org/10.1016/j.comnet.2016.08.017

1389-1286/© 2016 Elsevier B.V. All rights reserved.

As a new form of distributed systems for large-scale data pro-

essing, Cloud computing has been quite successful in integrating

everal applications areas including on-demand data centers [1] ,

emote processing and storage [2] , big data analytics [3] , and re-

ource virtualization [4] , offering different categories of services to

nd-users. Basically, the emergence of Cloud computing comes as

result of the significant evolution of high-speed networks, cou-

led with the wide-spread usage of Service-Oriented Architectures

SOA) that allows providing any kind of resources as a “service ” for

achine-to-machine (M2M) interaction. In the Internet world, web

ervices represent the major implementation of SOA with differ-

nt and complementary approaches including Simple Object Access

rotocol (SOAP) and Representational State Transfer (REST).

In the last four years, there has been an increasing interest for

ntegrating the Cloud facilities in cyber-physical systems [5] , which

as led to the emergence of new fields such as Cloud Robotics

6–8] , Sensor Clouds [9–11] , and Vechicular Clouds [12,13] . Re-

ently, a number of projects involving CPS have been conducted in

reas such as agriculture, food processing, industry, environmental

onitoring, security surveillance, and others. Meanwhile the num-

er of publications related to cloud CPS has been growing rapidly.

he authors conducted an extensive literature review by analyzing

elevant research papers in order to help researchers understand

he current status and future opportunities related to research in

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 261

Fig. 1. Frequency of publications in cloud CPSs (2011–2015).

c

i

i

i

n

e

h

f

c

r

b

m

a

t

t

c

d

d

c

t

v

t

b

T

i

s

s

a

f

h

p

a

m

t

w

a

l

t

h

c

c

o

o

p

s

T

o

i

2

f

a

t

m

2

t

m

s

t

c

r

o

r

n

m

s

d

c

h

s

p

i

n

s

t

c

a

c

loud CPS. This effort is focused on identifying the diversity of ex-

sting applications as well as proposed frameworks on cloud CPS

n industrial systems as well as smart home environment utilizing

ndustrial and/or service CPSs. Fig. 1 shows the trend of increase in

umber of publications listed in research databases including IEEE

xplore and science direct. The data collected from these databases

ighlights the number of journal as well as conference publications

rom years 2011–2015.

There are three main reasons for integrating CPS with the

loud. First, CPS devices, such as embedded sensor nodes or mobile

obots, are typically resource-constrained devices with limited on-

oard processing and storage capacities. Even though some CPSs

ight have appropriate resources, it induces a much higher cost

nd makes them unaffordable for large-scale uses. This results in

he need for offloading intensive computations from the CPS devices

o much more powerful machines in the cloud, where very high

omputing capacity is available at low cost. Second, the amount of

ata generated and used by CPSs is inherently very large as CPS

evices are in continuous and direct interaction with the physi-

al world getting instantaneous data to feed the computing sys-

em. Analyzing and extracting useful information from such large

olumes of data require powerful computing resources. In addition

o the need of offloading computation, abundant storage resources

ecomes a major requirement to cope with big data manipulation.

hird, cyber-physical systems are heterogeneous in nature, making

nteroperability a serious challenge. Accessing these heterogeneous

ystems from the Internet is not straightforward without having

tandard and common interfaces. To overcome this issue, there is

need to virtualize their resources and expose them as services to

acilitate their integration.

Considering the aforementioned challenges, it is clear that CPSs

ave three main requirements namely, ( i ) offloading intensive com-

utation, ( ii ) storing and analyzing large amount of data, ( iii ) en-

bling seamless access through virtual interfaces. These are the

ajor reasons why cloud computing represents a promising solu-

ion that addresses all these three challenges because, if coupled

ith a high-bandwidth connection to the CPS, it provides storage

nd computing resources as services to the back-end CPS, and al-

ows for easier integration of these CPSs by virtualizing the access

o them through standard web service interfaces. Fig. 2 explains

ow cloud computing can help to make cyber-physical systems ac-

essible and more powerful.

In this survey paper, we first give a motivation on how cloud

omputing can enhance CPSs. Next, we give in Section 3 an

verview on cloud computing. Then, we present a comprehensive

verview of latest research works on the integration of cloud com-

uting into three CPS areas namely, robotics ( Section 5 ), wireless

ensor networks (Section 5 ) and vehicular networks ( Section 7 ).

he integration is categorized along three axes: (1) computation

ffloading (or remote brain), (2) big data analytics, and (3) virtual-

zation.

. Motivation

The promising application of cyber-physical systems raised new

unctional requirements including real-time processing, storage,

nd accessibility and non functional requirements like security. In

his research work, we are interested in the functional require-

ents.

.1. Real-time processing

With the recent technological developments, cyber-physical sys-

ems can provide finished products and high quality services. This

ay pose the problem of executing many useful programs. For in-

tance, a mobile nursing robot [14] should autonomously achieve

he quality of care that nursing staff can provide. From a medi-

al informatics point of view, this kind of robot should read envi-

onmental data, perform intelligent navigation which may include

bject detection and/or tracking, support a robust and safe human-

obot interaction, etc.

While moving on the road, vehicles are also offering a certain

umber of applications and services simultaneously to achieve hu-

an safety and satisfaction [15] . Offering an acceptable level of

afety requires the execution of cooperative collision warning, inci-

ent management, and emergency video streaming. Besides, vehi-

les should execute a set of basic applications like platooning, ve-

icle tracking, and notification services. Drivers also may look for

ome comfort applications like parking place management, peer to

eer applications, and entertainment.

Therefore, cyber-physical systems should be capable of execut-

ng a variety of applications completely independently. Unfortu-

ately, the design of these systems usually imposes stringent con-

traints on computing performance. The limited resources, such as

he memory sizes, the battery capacity, and the processor speed,

annot satisfy the demands for executing these variety of complex

pplications.

Many research works have faced the problem of real-time pro-

essing in cyber-physical systems. As a matter of fact, authors in

262 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

Fig. 2. Cyber-physical systems clouds.

e

m

t

m

f

q

b

r

c

2

t

i

u

t

p

c

h

m

t

c

i

p

a

c

m

t

a

a

i

t

h

i

[16] designed and developed an IoT cloud platform to connect

smart devices (e.g. sensor nodes, robots) to cloud services for real-

time processing of data and control. In particular, they considered

the deployment of robotics applications with strict processing de-

lay and scalability requirements. They were motivated by the fact,

that for some resource constrained devices like sensors and robots,

only a small portion of data can be processed locally and the re-

mainder should be offloaded to a central server. A Roomba vacuum

cleaner robot was provided as an example. Experimental studies

involved a Turtlebot robot with a line follower application and

performance evaluation study demonstrated that it is possible to

achieve low delays with commodity hardware in the cloud and

it is possible to scale the architecture to hundreds of connected

devices.

Another example that needs real-time processing in sensor net-

works and Internet of Things is IoT analytics. In [17] , the authors

introduced a new approach to IoT data stream analytics for real-

time data acquisition, annotation and processing of sensor data.

The paper uses the OpenIoT platform of an EU FP7 project to de-

velop a framework for offloading data from sensor networks for

real-time data analytics in IoT applications. The server deploys

complex clustering algorithms to analyze data in real-time. Such

a tool is valuable for understanding complex phenomena such as

impact of air pollution in human health.

2.2. Storage

The rapid growth of the Internet of things with the latest ad-

vances of Information Technology (IT) have increased the amount

of data generated from cyber-physical systems. These systems are

now all over the world collecting data. Smarter vehicle manage-

ment, home automation systems, surveillance, etc are all applica-

tions where cyber-physical systems generate a massive amount of

data that should be stored somewhere for analysis.

The exponential growth in the amount of data generated

by cyber-physical systems needs a revolutionary storage strat-

gy. Memory or/and storage systems are fundamental performance

etrics and energy bottleneck in these systems [18,19] , in addi-

ion to communication. In the design of cyber-physical systems

emory issues play a key role, and significantly impact their per-

ormance. This memory capacity must scale with applications re-

uirements in terms of storage. Unfortunately, this scaling exacer-

ate the weakness of storage capabilities in these systems. This

epresents an additional reason for the need to offload data and

omputation to the cloud.

.3. Accessibility

The confluence of advances in wireless communication has led

o the mobility of cyber-physical systems. They became deployed

n applications where mobility is a necessity. For example, vehic-

lar networks are highly mobile networks as vehicles are consis-

ently moving, which induces additional challenges with respect to

rotocol design and performance issues. In [20] , the authors dis-

ussed the state of the art of IP mobility support solutions for ve-

icular networks to ensure their accessibility considering the two

ajor kinds of mobile communication models, namely the vehicle-

o-vehicle (V2V) communication and vehicle-to-infrastructure (V2I)

ommunication.

The major challenge with mobile CPSs is that they might be

naccessible either permanently or intermittently if mobility sup-

ort is not carefully addressed. Handoff and handover mechanisms

re typically integrated in mobile network communication proto-

ols to deal with mobility. In [21] , the authors proposed a handoff

echanism for low-power wireless sensor networks to maintain

he accessibility of sensors and evaluated its performance through

probabilistic model. The paper concludes that the handoff mech-

nism performs well in the transitional region of the link qual-

ty [22] , and may achieve up to 98% of throughput with delays in

he order of tenths of seconds. The use of cloud computing would

elp improve the effectiveness of these applications with mobil-

ty requirements. As an example, in [23] , the authors proposed a

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 263

m

c

a

c

a

n

c

r

3

(

v

c

s

m

t

a

e

t

M

a

p

i

(

v

s

l

m

e

S

v

i

C

v

a

o

v

I

s

t

r

v

c

f

m

z

a

g

A

d

a

4

c

r

i

a

i

i

I

t

r

1

[

r

t

t

s

l

obility-aware trust worthy crowdsourcing framework in a cloud-

entric IoT architecture. The framework incorporates user-mobility

wareness while considering malicious alteration of data. They

ame to the conclusion that mobility-awareness and reputation-

wareness in crowdsourcing IoT application improve the effective-

ess of smart city management authority.

Most of the limitations identified above will be remedied with

loud computing. In the following sections, we will discuss how

esearchers addressed these challenges.

. Cloud computing in a nutshell

According to the National Institute of Standards and Technology

NIST) [24] , cloud computing is a model for enabling ubiquitous, con-

enient, on-demand network access to a shared pool of configurable

omputing resources (e.g., networks, servers, storage, applications, and

ervices) that can be rapidly provisioned and released with minimal

anagement effort or service provider interaction . A simpler defini-

ion of cloud computing is having access to all your applications

nd data from any network-connected device. This approach is not

ntirely new since the concept of delivering computing resources

hrough a global network was introduced in the sixties by John

cCarthy who proposed the idea of computation being delivered

s a public utility [25] . With the evolution of Web 2.0, cloud com-

uting has evolved through several phases including grid and util-

ty computing until the delivery of the first Software as a Service

SaaS) with Amazon EC2/S3.

The service model in cloud computing defines the type of ser-

ices you can access in the cloud. Three service models are univer-

ally accepted: (1) Infrastructure as a Service (IaaS), where the de-

ivery of hardware infrastructure over the network on pay by use

odel, (2) Platform as a Service (PaaS), which provides a software

nvironment where consumers can deploy their applications, (3)

oftware as a Service (SaaS), which offers a complete operating en-

ironment that contains management, applications, and the user

nterface.

These services can be deployed in four models: (1) Public

louds : cloud services in public clouds are made available by a ser-

ice provider for open use by the general public. It is owned by

third-party selling cloud services like Amazon AWS [26] , which

wns the infrastructure and offers the access via Internet. (2) Pri-

ate Clouds : are built for exclusive use of single client/organization.

t may be managed by the organization itself or a third party. De-

pite being an expensive solution, the private cloud basically offers

he highest level of security and control. (3) Hybrid Clouds : rep-

esent a combination of two or more distinct clouds to accomplish

arious functionalities within the same organization. For instance a

ollaboration between VMware [27] and Google cloud [28] has of-

ered an hybrid cloud solution called VMware vCloud Air. (4) Com-

unity Clouds : represent a collaboration between several organi-

ations to serve a common function or purpose. It may be oper-

ted, owned, and managed by a third party or the constituent or-

anization(s). A good example of community cloud deployment is

pps.gov [29] which provides cloud services to federal agencies.

The definition of cloud computing requires that the practical

esign of cloud system must consider the following five keys char-

cteristics:

• Scalability: the cloud needs to dynamically increase the re-

sources to serve the increasing workload as and when required.

Scalability in cloud computing means: (1) the ability of an ap-

plication/product to function well even when it increases in

volume or size to meet the user requirement, and (2) taking

full advantage of the rescaled situation.

• Usability: cloud computing services are supposed to serve users

from different domains. To achieve clients satisfaction, the ac-

cess to cloud should be as simple as possible by designing a

comprehensive and a user-friendly interface to cloud.

• Reliability: it refers to the probability of a system, including all

of its hardware and software components, to perform correctly

as expected. With third dependencies on remote services and

databases, network connection and hardware cloud services are

downright complicated. As such, reliability of the cloud is a ma-

jor concern considering its distributed and multi-tenancy na-

ture. In [30] , the authors discussed security and reliability is-

sues in cloud computing systems. A cloud system should pro-

vide the services it is designed for with no major flaws and this

induces additional challenges to their design and development.

Verification, validation and testing techniques should be used

to ensure reliability of these systems.

• Security: when you consider moving your application to the

cloud, consumers must understand the risks and security as-

sociated with the cloud. They should ensure the privacy of

their sensitive data [31] . Security and privacy presents a lot of

challenges in todays cloud computing systems, like identifying

new threats and vulnerabilities, protecting virtual infrastruc-

tures, protecting outsources of computation and services, secur-

ing big data storage and protecting user data [32–34] . The use

of integrated security solutions at different levels that span over

authentication, encryption, intrusion detection and prevention

systems, trust management, etc., should be carefully designed

based on the application requirements.

• Cost: the cost is one of the most important factors regarding

the use of cloud computing platforms. It is generally based

on the concept pay-as-you-go taking advantages of the elastic

computing capabilities provided by the cloud. However, decid-

ing whether the cost of subscribing to cloud computing facili-

ties or building its own data center is a highly complex prob-

lem that depends on several factors, as lower cost is not al-

ways guaranteed when using cloud services. A cost effective-

ness analysis should be conducted when cloud computing plat-

form is to be used [35] .

. Cloud robotics

Nowadays, we are at the cusp of a revolution in robotics. The

urrent research trend aims at broadening the usage of service

obot systems as well as industrial robots in various applications

ncluding smart home environments [36] , smart office buildings,

irports [37] , shopping malls [38] , manufacturing labs, integrated

ndustrial control systems, etc. The demand for these service robots

s increasing in a world with significant attachment to the use of

nformation and Communication Technologies (ICT). The Interna-

ional Federation of Robotics reported that the number of service

obots for personal and domestic use sold in 2014 increased by

1.5% compared to 2013, augmenting sales up to USD 3.77 billion

39] . It is also forecasted that the sale of household robots could

each 25.2 million units during next three years.

Although service robots have been useful for several applica-

ions and various uses, their use is limited comparatively to other

echnologies such as smartphones, mobile phones, tablets, etc. Es-

entially, robots have limited exposure to the general public at a

arge scale for the following reasons:

• Robots have been typically used as standalone systems for very

specific and dedicated missions in controlled environments,

such as in industrial manufacturing [40] , hospitals [41] etc.

Even in the case of cooperative and multi-robots applications,

robots communicate with users or other robots to perform spe-

cific pre-programmed, pre-defined missions, yet are isolated

from the external world. Consequently, they cannot learn from

their context.

264 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

r

t

s

T

t

a

t

f

e

t

m

z

t

R

b

m

r

t

c

e

r

r

p

p

c

a

c

t

[

i

c

r

o

t

c

fl

t

c

r

l

h

c

l

a

w

m

a

a

t

a

a

t

n

v

R

q

M

T

c

g

t

o

s

• The complexity of configuration and maintenance of the

robots make their use challenging for non-technical and non

computer-savvy users.

• The relatively high cost of sophisticated service robots can be

cost abhorrent for public users which confined their utilization

in professional organizations or specialized domestic applica-

tions.

Given the presence of the aforementioned restrictions, there is

an increasing demand and interest in using the Internet infrastruc-

ture as a means of promoting robotics applications from two per-

spectives; by exposing robot resources as Internet services to end-

users; and by enabling the robot-to-robot (R2R) as well as robots-

to-end users communications through the Internet.

With the emergence of cloud computing, robots may also reap

benefits from the huge computing resources available in the In-

ternet to increase their processing capabilities for computation in-

tensive applications. The cloud robotics paradigm, coined in 2010

by James Kuffner, seems to be the missing building block to-

wards jumping to a new frontier of public use of robots. Indeed,

cloud robotics has opened several new research ideas and interest

with respect to coupling the Internet and cloud computing with

robotics.

Cloud computing leverages the limitations of embedded hard-

ware by enabling the robots to offload computation and mas-

sive storage requirements to the remote cloud infrastructure which

would be used as a “remote brain”, where all the computation

would be done based on the sensor data accumulated from the

robot. Furthermore, advances in networking technologies and high

speed mobile networks have enabled fluid interaction between the

cloud and the robots. Benefits of aforementioned enabling tech-

nologies have unveiled the possibility of sharing and collaboration

among robots, machines, smart objects and humans.

Cloud robotics evolved over the last few years from the con-

cept of networked robotics. With the availability of the world wide

web, online web robotics became a reality. In the nineties, some

of the earliest work by K. Goldberg et al. [42] [43] resulted in

web based tele-operated robots controlled remotely via an Internet

browser. In [44] , McKee described the benefits of distributed pro-

cessing in networked robotics and challenges in communication,

synchronization and performance. In 2010, Kuffner [45] coined the

term, “Cloud Robotics” in his paper on cloud-enabled robotics, pro-

viding a wide vision where robots can harness the massively par-

allel computation resources and vast data storage available via the

Internet.

In this section, we will present three facets of cloud robotics:

(1) Remote Brain: Offloading Computation, (2) Remote Storage: Big

Data Processing, and (3) Remote and Distributed Collaboration.

4.1. Offloading computation

Nowadays, applications requiring massive computations can be

remotely accomplished in the massively parallel cloud infrastruc-

ture. It has become possible for mobile and service robots to over-

come the limited onboard resources by “offloading” massive com-

putation tasks on remote servers via the Internet. Consequently, in

the near future, robots would be used as actuating devices, captur-

ing and offloading sensor data to the cloud infrastructure for rapid

processing and analysis. Upon completion of the processed tasks,

robots would actuate accordingly, driven by the response received

from the cloud.

Robots are typically capable of processing information using

multiple onboard processors. However, certain types of robots such

as mobile robots or Unmanned Aerial Vehicles (UAV) may have

limited on-board resources lending to stringent dimensional size

and limited payload requirements. As an example, surveillance

obots [46] require performing real time mapping, localization, au-

onomous navigation and object recognition. The robot can capture

ensory data from its environment and send it to a cloud service.

he cloud service permits running interactive, real-time processing

asks in parallel, including applications for video surveillance, im-

ge building using Point Cloud libraries (PCL), sensor information

racking, predicting and evaluating performance and other tasks.

Recently, researchers have addressed the concept of “offloading”

rom robots to cloud infrastructure. Rapyuta proposed by Hunziker

t al. in [47] is the ROBOEARTH cloud engine devised to overcome

he challenge of managing robotic resources constraints by imple-

enting an open source cloud robotics framework. It uses Ama-

on data center to allow outsourcing of robots on-board compu-

ational processes by providing secure, elastic, customizable, and

OS compatible computing environment in the cloud. This ground

reaking work presents opportunities for a Software as a Service

odel that may allow custom applications to offload data from

obots to the cloud service, which can process the data and re-

urn outputs. This allows the users to avoid the burden of pro-

essing and managing data onboard the device. In [48] L. Bingwei

t al. present Cloud Enabled Robotics System (CERS) that allows

obots to mesh together to form a local networked system. These

obots can offload heavy computation tasks to the server side im-

lementation of a cloud enabled software architecture. The pro-

osed system also considers security implications while merging

loud and robot networks. Researchers tested the system for an

pplication of real-time video tracking and provided performance

omparison based on virtual machines and physical machine clus-

ers. Another example of computation offloading is presented in

49] where a cloud robotic system with real time face recognition

s implemented. Due to the resource-hungry face recognition appli-

ations requiring complex computations from resource-constrained

obotic systems, researchers exploited the computational services

f the cloud by offloading heavy computational tasks from robots

o cloud servers.

Turnbull in [50] developed a full scale cloud infrastructure to

ontrol the behavior and formation of a multi-robot system by of-

oading computation tasks to the cloud. Robots within the sys-

em are designed with minimal hardware, equipped with a micro-

ontroller with WiFi capabilities for robot-to-robot (R2R) as well as

obot-to-cloud (R2C) communications. All the computational work-

oad and execution of algorithms required to control the robot be-

avior are performed in the cloud. The cloud infrastructure re-

eives data from a vision acquisition system, determines the robot

ocation and status, implements algorithms to control its behavior,

nd then sends appropriate commands to the robot. The system

as validated using iRobot Create equipped with LinkSprite Dia-

ondback microcontroller, a cloud infrastructure using XenServer

s the hypervisor and XenCenter as the virtualization manager, and

single virtual machine with Ubuntu 12.10 and MATLAB installed.

Authors in [51] have exploited the Service Oriented Architec-

ure (SOA) and cloud computing technologies to perform robotic

pplications like path planning, object recognition, map building

nd navigation in the cloud using the Map Reduce computing clus-

er. The designed framework is composed of three main compo-

ents: (1) Cloud manager which receives incoming service requests

ia http/https and checks the authorization of requesting client, (2)

obotic service handler to execute and check the availability of re-

uested robotic applications exposed as robotic services, and (3)

ap Reduce computing cluster to process large amount of data.

he authors have demonstrated the effectiveness of Map Reduce

omputing cluster by performing a real time speech based navi-

ation of create robot using VPL of MRDS to simulate and control

he robot. The authors discuss in detail the performance aspects

f execution of map-reduce on the cluster. Much needs to be de-

ired in terms of validation of the architecture under execution of

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 265

r

f

i

a

F

v

t

t

t

c

F

v

r

b

t

t

t

t

b

n

o

p

c

t

h

f

a

c

t

t

p

t

w

f

o

t

i

t

t

d

l

m

c

c

t

o

c

b

4

s

t

r

s

(

b

w

h

y

r

k

p

l

p

s

m

h

r

k

G

t

p

b

r

t

t

H

u

b

r

t

a

f

d

R

b

e

k

h

E

m

W

o

a

R

p

m

w

m

t

a

w

s

G

b

o

u

r

4

s

r

o

a

p

a

o

u

f

f

n

f

f

eal-life applications scenarios. Another application of Map-reduce

ramework executing on Hadoop is the DAvinCi project reported

n [52] . The researchers provide the DAvinCi server that acts as

proxy server coupling the two eco-systems, Hadoop distributed

ile System (HDFS) and Robot Operating Systems (ROS). The ser-

ice provides a publish/subscribe model for robots to send/update

heir information as well as requests, which are logged and main-

ained on the server. The server pushes this data and/or requests

o the backend HDFS. The server triggers MapReduce tasks to exe-

ute data and collects results produced by the Hadoop ecosystem.

inally these results are transmitted to the ROS subscribers. They

alidated this work by implementing the FastSLAM algorithm and

eported significant performance gains in execution times provided

y the framework. Although the focus of the paper is the evalua-

ion of FastSLAM algorithms under large distributed environments,

he communication issues between internal processes as well as

he cloud is not detailed. We notice that the implementation of

he server component is dependent on reliable communications

etween the namenode in the hadoop cluster and the ROS master-

ode. Although hadoop provides backup measures by invoking sec-

ndary namenode in-case of the primary namenode failure, the

roposed framework lacks to address the reliability and communi-

ation latency or failure issues of the DAvinCi server between the

wo ecosystems. The researchers implemented the system using a

adoop cluster composed of only eight nodes. They plan to address

ailsafe mechanisms for communications between ROS ecosystem

nd the DAvinCi server.

R2R and R2C Communications are of utmost importance for

omputation offloading applications. To address the shorcomings of

he DAvinCi project, authors in [53] , developed a system for real

ime moving objects recognition and tracking using cloud com-

uting to perform offloading of heavy tasks. To guarantee real-

ime performance, they developed an offloading decision frame-

ork which estimates the computation and communication times

or tasks and decides to execute the tasks either on the robot

r the cloud server. The resulting decision minimizes the execu-

ion time while maximizing tasks completion rate while satisfy-

ng real-time constraints. While the authors performed computa-

ion offloading, they have not used cloud computing platforms or

echnologies, but rather a simple server machine. In addition, the

eployment is rather limited to a local area network with wire-

ess connection, and not tested on Internet infrastructures. Further-

ore, it is applied for a stationary robot, which limits its appli-

ability to such context. Another example of research in reliable

ommunications is presented in [49] . Researchers address the data

ransfer latency between the mobile robot and the cloud server in

rder to perform real time face recognition. They proposed an ar-

hitecture that supports high-speed data transmission and distri-

ution of computation load.

.2. Big data processing

Modern and future robots would be able to deal with large

cale sophisticated data including several types of sensor data, cap-

ured images etc. Thanks to recent advances in communications,

obots are increasingly becoming capable of capturing and posting

ensory data to cloud infrastructure, realizing Internet of Things

IoT) [54] . Given the inherent physical storage limitations for mo-

ile and service robots, providing access to remote cloud storage

ould allow robots to post sensor data, populate remote databases,

elp build and share knowledge bases, perform remote data anal-

sis without the need for additional onboard memory. The cur-

ent research trend is using data populated by robots to build

nowledge-bases for collective robot learning. This leads to im-

roved algorithms for robot path planning, object recognition, etc.,

eading to a collective intelligence space for robots to share.

Several knowledge databases for robots have recently been pro-

osed. Robo Brain [55] is a project created to build a shared, large-

cale knowledge base for robots to be able to function in hu-

an environment. This project helps robots to learn human’s be-

avior. The core idea of this project is to enable newly deployed

obots in an environment to connect and learn from the robo-brain

nowledge-base instead of beginning from scratch. The Columbia

rasp dataset [56] is a large dataset of pre-computed grasps on

housands of 3D models designed for learning and benchmarking

urposes. To construct this database, authors in [57] have com-

ined (1) RoboEarth database knowledge to give a description of

obot tasks with (2) the distributed task execution methods of

he Ubiquitous Network Robot Platform [58] to abstract the access

o robot hardware and software components. The Willow Garage

ousehold Objects Database [59] contains thousands of 3D models

sed to evaluate various aspects of grasping.

Google Goggles, [60] a free image recognition service for mo-

ile devices, has been incorporated into a cloud-based system for

obot grasping. Users can take a photo of an object and send it

o Google for analysis and storage. Authors in [61] have designed

knowledge base about robotics rehabilitation. The database in-

ormation is made available through web-interface allowing robots

esigners to update information about their robots rehabilitation.

EHABROBO-ONTO is presented in Web Ontology Language (WOL)

ased on the standards of World Wide Web Consortium (W3C) to

nable integration with medical ontologies. The concept of shared

nowledge database was also used with industrial robots to meet

uman’s intelligence when working in dynamic environments. The

uropean Union’s ROBOEARTH project [62] was a leader of the re-

ote memory idea. The main objective of this project is to build a

orld Wide Web for Robots for sharing knowledge about actions,

bjects, and environments between robots [63] . Robots can store

nd share information in a common presentation designed by the

oboEarth language.

Authors in [64] have designed a rehabilitation database using

hysical therapy robots. Therapy robots equipped with the required

aterial can record and send the data to a rehabilitation server

hich in turn saves the data and maintains the database. Perfor-

ance evaluation has proven that the results using the rehabilita-

ion database are acceptable compared to those generated by ther-

pists. However, these results remain theoretical as the database

as not tested on real patients. Ben Kehoe et al. in [65] have de-

igned an architecture in which they connect robots to the Google

oggles image recognition system. The robot sends data collected

y its depth sensor and camera to a Google server that performs

bject recognition and returns a set of grasps with confidence val-

es. A prototype of this architecture was developed and evaluation

esults showed comparable performance.

.3. Virtualization

Industrial and personal robotic systems are designed to provide

ervices to human beings in a variety of circumstances. Sets of

obots accessed remotely can collaborate together to perform vari-

us functions, tasks and sub-tasks to achieve a unified goal. There

re several issues and challenges that require attention: (1) robot

rogramming and manipulation is usually complex for beginners

nd naive users, (2) monitoring and controlling robots usually rely

n complex programming routines that cannot be handled by end-

sers, (3) robots are typically expensive platforms and are not af-

ordable for a large audience. In this context, researchers profited

rom cloud computing concepts and recent advances in web tech-

ologies to virtualize robotic systems, hiding complexities and of-

ering robot-as-a-service to the end-users.

Authors in [66] present Jeeves, a distributed service framework

or cloud based robot services. The key mechanism of the proposed

266 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

s

r

t

b

a

m

t

a

f

e

c

T

s

(

i

t

S

r

o

v

t

i

a

l

a

d

d

c

p

5

g

t

d

i

m

s

h

n

t

m

u

t

e

o

n

a

i

p

r

e

i

(

i

c

t

i

W

framework is a robot assignment function, which discovers dis-

tributed robot resources and assigns the requested tasks by end

users to suitable robots. Jeeves integrates various devices including

robots with Internet services and provides a platform for offload-

ing and executing computations in the cloud. Despite authors have

proved the effectiveness of their architecture, they didn’t provide a

secure access to the cloud.

Authors in [67] have designed and implemented a cloud in-

frastructure that virtualizes robots and delegates tasks to robots.

When the system receives a robotic application to perform, it di-

vides the task into sub-tasks and assigns the sub-tasks to local

robots and robots that belong to other clouds. The authors defined

a robot node level virtualization that allows robots to run multi-

ple applications at the same time. The designed system comprises

three main components: a P2P overlay, a robotic cloud and a VRC

(Virtual Robotic Cloud). The P2P overlay is used as the interaction

network between different robotic clouds through the VRC which

is defined as a node that facilitates the communication between

robotic clouds via the overlay. The architecture model presented in

this work is a gateway-based model, therefore it is of high interop-

erability. However, the gateway is a failure point which may affect

the architecture robustness.

The Unmanned Aeriel Vehicles (UAVs) area also attracted re-

cent research works on SOA and REST Web services. In [68] au-

thors proposed a mapping between cloud computing resources and

UAVs resources based on a SOA model. They offer two kinds of

services: essential services (e.g. mission organization service, bro-

ker service, ground station commands, etc.) and customized ser-

vices (e.g. sensing services, actuation services, image/data analysis,

etc.). The paper only provides high-level description of system ar-

chitecture, components and services without any specific detail on

how these could be implemented on a real system. In addition, the

authors did not discuss performance metrics and tradeoffs related

to the use of UAVs in the cloud, and no quantitative performance

analysis was presented. In addition, scalability issues with the de-

ployment of multiple UAVs on the cloud were not discussed. In

[69] , The same authors improved their previous work and designed

a RESTful web services model by following a Resource-Oriented Ar-

chitecture (ROA) approach. They have also designed a broker that

dispatches mission requests to available UAVs. The broker is re-

sponsible for managing the UAVs, their missions and their inter-

actions with the client. A prototype for emulating the UAV and its

resources was implemented on an Adruino board. This prototype is

very limited as it does not demonstrate a sufficient proof of con-

cept on real drones or robots which doesnt effectively demonstrate

the feasibility of the approach.

In [70] , Koubaa proposed RoboWeb, a SOAP-based service-

oriented architecture that virtualizes robotic hardware and soft-

ware resources and exposes them as services through the Web.

RoboWeb consists of the integration of different Web services tech-

nologies with the ROS middleware to allow for different levels of

abstraction (multi-layer architecture). It was designed to develop a

remote lab which allows researchers and students to remotely ac-

cess and monitor robots that belong to the framework. This system

architecture comprises three main layers: (1) the web interface

layer which implements rosPHP library that provides ROS function-

alities allowing access to ROS-enabled robots, (2) the roboWeb ser-

vice broker which is defined as a middleware allowing interaction

between end-users and ROS-enabled robots, and (3) the robotic

ecosystem which comprises ROS-enabled robots including Tutlebot

and Wifibot. Researchers have validated their proposed architec-

ture by developing a web interface that allows non technical users

to access, control and monitor robots that belong to the frame-

work. Unfortunately, the architecture offers a low level of security.

Password-based authentication doesn’t adequately provide a strong

security.

In [71] , the authors proposed a remote robot lab for conducting

hared experiments and developments that allows the control of

obots remotely through a web interface. They made extensions to

he Robot Operating System (ROS) middleware using rosjs, and ros-

ridge, through web sockets accessible anywhere over the Internet,

nd provides security mechanisms and runtime tools for remotely

anipulating the robots. They also present two robotic experimen-

al environments that implement rosjs technology in order to visu-

lize and interact with complex robot platforms.

Rsi Research cloud proposed in [72] is a cloud computing plat-

orm that provides robotic services through the Internet. Authors

xploited the Robot Service Network Protocol (RSNP) to hide robots

omplexities and expose robotic services to non-experts in robots.

hey have tested their proposed architecture by implementing a

urveillance camera service.

Chen et al. in [73] defined the paradigm Robot as a Service

RaaS), proposing a cloud framework for interacting with robots

n the area of service oriented computing. The authors exploited

he SOA to design and implement a prototype of the Robot as a

ervice (RaaS) cloud computing model which consists of exposing

obotic services and applications and allows interaction with devel-

pers and end-users. The design complies with the common ser-

ice standards, development platforms, and execution infrastruc-

ure, following the Web 2.0 principles and participation. RaaS unit

s composed of three main parts: (1) service provider allowing to

dd/remove services to/from the robot, (2) service consumer al-

owing end-users and developers to benefit from deployed services

nd compose new applications, and (3) service broker for research,

iscovery and publishing of robotic services and applications in the

irectory.

In Table 1 , the architectures for integrating robots with cloud

omputing are compared with respect to the key requirements

resented and discussed in Section 3 .

. Sensor clouds

WSNs [74] are considered as one of the enabling CPS technolo-

ies in the 21st century [75] and one of the major contributors to

he IoT. WSNs currently represent an invincible technology embed-

ed into a wide variety of industrial and CPS applications, includ-

ng healthcare [76] , environmental monitoring [77] , defense and

ilitary [78] , industrial automation [79] , smart environments [80] ,

mart grids [81] , to name a few.

Over the last fifteen years of WSNs research, the major focus

as been on the design of lower layer protocols including commu-

ication, medium access control, networking, topology control pro-

ocols and these reached a certain saturation. In addition, with the

aturity of the field, the current trend nowadays is to adopt the

se of standard communications and networking protocols rather

han devising proprietary solutions. Very recently, and with the

mergence of the IoT paradigm, the focus is switched to be more

n the integration of WSNs into the Internet, which raises several

ew challenges, including compatibility and interoperability issues,

nd interfacing between the WSNs world and the Internet. The first

ntegration step came with the design of the IPv6 version for low-

ower networks, namely the 6LoWPAN, which enabled the flow to

un between wireless sensor devices and the IP World. Later, sev-

ral other networking protocols emerged such as RPL (IPv6 Rout-

ng Protocol for Low-Power and Lossy Networks) [82] and LOADng

6LoWPAN Ad Hoc On-Demand Distance Vector Routing) [83] rout-

ng protocols. The second step was the integration at the appli-

ation layer by enabling IP network users to access sensor data

hrough Web interfaces. SOA and Web Services [84] [85] played an

mportant role into this integration, in particular the use of RESTful

eb Services as they are lightweight and can be more easily sup-

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 267

Table 1

A summary of cloud robotics architectures.

Work Scalability Reliability Cost Usability Security

Rapyuta platform [47] yes yes yes – yes

Cloud enabled robotics system (CERS) [48] yes – – yes yes

Cloud robotics: formation control of a multi robot system utilizing cloud infrastructure [50] yes no yes – no

Robotic services in cloud computing paradigm [51] yes no no yes yes

DAvinCi project [52] yes no no yes no

Cloud robot with real-time face recognition ability [49] – – – yes no

Real-time moving object recognition and tracking using computation offloading [53] yes yes – – no

Cloud-based robot grasping with the Google object recognition engine [65] yes no no – no

An implementation of a distributed service framework for cloud-based robot services [66] yes – yes yes no

An infrastructure for robotic applications as cloud computing services [67] yes no – no no

A service-oriented architecture for virtualizing robots in robot-as-a-service clouds [70] yes – yes yes yes

Broker architecture for collaborative UAVs cloud computing [69] yes – – yes yes

rosjs [71] yes – yes yes yes

Research and development environments for robot services and its implementation [72] yes – no yes no

Robot as a service in cloud computing [73] yes – yes – no

p

W

g

r

r

t

a

e

b

g

T

a

m

b

t

p

d

m

p

e

e

e

W

d

s

t

o

O

5

w

r

c

s

c

i

e

c

r

h

i

l

I

fl

f

p

i

n

t

f

T

t

c

(

a

d

t

o

p

h

a

fi

t

W

c

d

p

f

c

p

t

p

r

c

f

c

n

w

t

d

n

c

w

c

[

t

a

a

r

orted by low-power devices such WSNs as compared the SOAP

eb Services.

All these effort s have paved the way to a more consistent inte-

ration of WSNs with the Internet using the cloud paradigm and

esulting in what is called Sensor Clouds. Considering the limited

esources of typical wireless sensor nodes, their integration with

he cloud is an important milestone to leverage WSN applications

nd promote their performance and capabilities. In fact, with the

mergence of the Internet-of-Things (IoT), wireless sensors are em-

edded nowadays in trillions of devices around the globe and keep

enerating data in real-time from the surrounding environment.

ypically, sensor-enabled devices are merely data generators and

re too restricted to perform extensive operations to analyze and

ake use of the data they generate. However, their wireless capa-

ilities allow them to transfer their data to more powerful devices

o make use of this data. It is here where the cloud comes into

lay by providing different computing and storage resources, in ad-

ition to a wide array of applications and utilities to process and

anipulate the big data. Furthermore, users may need to access

articular sensor devices to read data or modify its settings. For an

ffective user-sensor interaction, interfaces and management lay-

rs are required. The cloud provides such facilities through differ-

nt technologies centered around service oriented architecture and

eb services.

There has been several research effort s towards the design and

eployment of sensor clouds [86] to offer a remote sensor data

torage, management and integration. In what follows, we present

he major contributions in the field, and discuss the opportunities

ffered by sensor clouds in the three mentioned perspectives (1)

ffloading Computation, (2) Big Data, and (3) Virtualization.

.1. Offloading computation

With the increasing popularity of WSN in various domains, a

ide range of potential and promising applications is evolving and

equires processing large amount of sensor data in real-time to

ope with the needs of complex information processing algorithms

uch as object detection and tracking, sensor-based surveillance,

ontrol and automation, etc. These applications are quite demand-

ng in terms of processing and storage and require the use of pow-

rful and scalable processing infrastructures to provide the needed

omputation services to low-capability sensor devices.

The current research trend aims at harnessing cloud computing

esources to augment resources capabilities of WSNs by migrating

eavy processing applications to the cloud, which save energy and

mprove their performance.

A significant amount of research has been performed on off-

oading computation to the cloud. With the emergence of the

oT paradigm, ANGELS [87] was proposed as a framework to of-

oad huge volume of data and to provide a parallel processing in-

rastructure among smart devices surrounding us, such as mobile

hones. The key idea was to use mobile devices when they are

dle to execute programs and process data, thus contributing to a

ew paradigm of distributed computing in the IoT. The cloud archi-

ecture consists of a server machine with a HADOOP Map/Reduce

ramework which mediates between end-devices for job offlaoding.

he server module comprises three components: (1) the Data Parti-

ioner , which partitions input data according to the score of devices

alculated using AndEBench benchmarking tool for android devices,

2) the Job Scheduler , which maps the data partitions to the nodes,

nd (3) the Combiner which combines results received from mobile

evices and produces the final result. The authors have experimen-

ally validated their proposed framework by estimating the value

f π using four android mobile phones. Although still in an ex-

erimental phase, ANGELS framework provides a new insight on

ow low power devices like smart phones could be used behind

cloud for remote and distributed processing. However, the ef-

ciency of such a system remains an open question considering

hat these mobile devices cannot handle extensive computations.

e note that this approach might seem to contradict the general

oncept that computation should be offloaded from low-capability

evices such as smartphone to a cloud infrastructure with higher

erformance commodity hardware. However, in general sense, it

ollows the same concept, as in the ANGELS context, computation

an indeed be offloaded to smartphones, which might have low

rocessing and storage capabilities, but their integration behind

he cloud may provide sufficient resources for some kinds of ap-

lications like those considered in ANGELS.

Authors in [88] tackled the problem of the limited processing

esources in WSNs. They proposed a component-based model ar-

hitecture that integrates WSNs with IaaS Hybrid Cloud to allow

or real-time data transmission and processing. The architecture is

omposed of three main tiers: (1) the sensor tier which uses a sink

ode to collect and send sensor data to a local proxy, (2) the gate-

ay tier that consists of a local proxy that uses LooCI middleware

o dynamically generate, register and control components for each

etected sensor node. Then, the local proxy transmits generated

ode components to the cloud and send sensor data to the related

omponent for processing via the Event Bus of the LooCI middle-

are, and (3) the cloud tier which consists of a hybrid cloud that

ombines Eucalyptus Private Cloud [89] and Amazon Web Services

26] . The authors experimentally proved that the proposed archi-

ecture induces only a very little overhead in terms of latency with

n average of 0.6 ms for the invocation of dynamic components. In

ddition, it minimizes the memory footprint on sensor devices and

educes energy consumption.

268 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

w

t

v

v

a

b

p

m

s

c

p

t

f

a

t

t

v

l

p

t

e

c

p

t

t

t

(

b

m

m

c

w

t

d

c

f

p

l

f

a

h

s

s

e

p

h

(

i

5

b

d

a

a

a

m

s

t

e

s

l

e

In [90] the authors proposed a cloud-based mobile healthcare

system, which enables end-users to control, monitor and improve

their health using mobile devices. The limitations of the mobile de-

vices are avoided by migrating computationally intensive operation

to cloud servers. Two multi-cloud offloading approaches were pro-

posed. The first model is a self-reliant architecture, which intends

to promote stability at the cost of a higher communication over-

head; while the second is called multi-cloud offloading, which pro-

vides a better tradeoff between stability and communication cost.

However, the authors do not give any tangible implementation of

such a system, and only provides an analytical study on the per-

formance of both approaches.

In [91] , the cloud was used to offload GPS signals to be pro-

cessed by the cloud instead of being processed by the mobile

device. The main motivation behind this was to reduce energy

consumption of mobile phones. The key idea consists of exploit-

ing cloud computing resources to generate a number of candidate

landmarks, and then use other geographical constraints to discard

the wrong solutions. It was experimentally proven that using the

proposed Cloud-Offloaded GPS (CO-GPS) solution, sensing a GPS lo-

cation takes 3 orders of magnitude less energy than GPS on mobile

phones. While this solution was shown to be effective in terms of

energy efficiency, it has some limitations with respect to localiza-

tion accuracy due to accumulated time drifts.

5.2. Big data processing

With the world-wide availability of low-cost and smart sensors

wirelessly connected, sensors are being embedded in all sorts of

devices of a variety of application domains including health, indus-

try, environment, and household, enabling automation and making

human life easier. Many organizations like IBM, The Central Ner-

vous System for the Earth (CeNSE), and The Wireless World Re-

search organization predict the growth of sensors to reach trillions

by 2020. Such enormous number of sensors produce huge amount

of data, which is growing exponentially and is expected to reach

40 zettabytes in 2020 [92] . In addition, as sensors are used in dif-

ferent areas and applications, they are sources of heterogeneous

and diverse types of collected data, which can be either structured ,

such as temperature and pressure, or unstructured such as images,

videos and audios. This results in the concept of big data that pro-

vides new opportunities, yet challenges, in what concerns process-

ing and analysis to extract useful information. Indeed, WSNs pro-

vide useful big data that can be analyzed and processed to serve

human life in several areas. For example, agriculture field [93] is

complex and requires long-time experiences to make critical deci-

sions such as predicting the yield of crops, which depends on en-

vironmental condition data and plant growth data. Researchers in

[94] discussed the importance of big data in business field by al-

lowing to make critical decisions, reduce risks, predict future out-

comes and save money.

In the literature, several recent research works addressed the

use of cloud computing technology to tackle big data challenges re-

lated to WSNs. In [95] , the authors proposed a scalable and secure

cloud-based architecture to collect, store and share huge amount

of confidential data generated by medical body sensor networks

allowing the healthcare institution to monitor patients’ health and

improve rescue processes in emergency cases. Authors in [96] pro-

posed a sensor-enabled patient monitoring system based on cloud

computing concepts to enable remote monitoring of patients in or-

der to reduce hospitals congestion. The proposed architecture is

composed of three main servers including (1) patient information

server or storage server which is responsible for storing terabytes

of sensed data, (2) medical server and (3) communication server ,

which are designed to analyze the gathered data, detect and dis-

cover anomalies, and provide caretakers and medical specialists

ith real-time information about patients’ health status. Authen-

icated healthcare professionals and patients can consult this data

ia the Internet. The feasibility of the proposed architecture was

alidated with two scenarios including monitoring elderly people,

nd post-operative patients. In [97] , the authors proposed a cloud-

ased architecture that automates the collection and processing of

atient’s vital data and provides a real-time remote access to the

edical staff. The wireless sensor nodes planted in patient’s bed-

ide collect data and send it to the cloud to be stored and pro-

essed on virtual machines managed by OpenNebula cloud com-

uting platform. The authors concluded from preliminary work

hat the architecture provides good performance. However, no per-

ormance evaluation were conducted to prove the efficiency of the

rchitecture.

Researchers in [98] have designed and implemented an archi-

ecture that uses Web Services and cloud computing technologies

o handle big data collected from WSNs. In [99] , the authors sur-

eyed existing tools and platforms for gathering and transmitting

arge volumes of sensor data to clouds such as MicroStrains, Tem-

oDB, SensaTrack Monitor and Ostia Portus platforms. They iden-

ified a set of research challenges namely, communication, data

xchange formats, security and interoperability. To address these

hallenges, the authors proposed a cloud-based sensor monitoring

latform to collect, store, analyze and process big sensor data, and

hen send control information to actuators. The proposed architec-

ure depicted in Fig. 3 uses the Generic Cloud Interface as an in-

ermediate layer between sensor servers and the cloud to assure:

1) secure data access from and transfer to the cloud, (2) porta-

ility and interoperability, and (3) efficient data transport by for-

atting the sensor data into platform-neutral data exchange for-

at, namely XML and JSON. In [100] , researchers addressed the

loud-enabled large-scale WSNs aiming at integrating sensor net-

orks with cloud infrastructures to manage Big Data. They used

he Apache Hadoop framework to store and process big sensors

ata. In [101] , the authors presented their vision for data quality

entric big data infrastructure for federated sensor service clouds

or what concerns collection, management and analysis. They have

roposed the design of a cloud-based big data architecture to al-

ow gathering, analyzing, storing and sharing large volume of data

rom large numbers of heterogeneous sensors. They also discussed

pricing model and Service Level Agreements (SLAs). Researchers

arnessed big data technologies for storage and analysis of sen-

or massive feeds. They took benefit from: (1) NoSQL document

tore and key-value storage technologies to provide a scalable and

fficient storage of historical sensor fed big data, and (2) Stream

rocessing and Map-Reduce technologies for analyzing online and

istorical sensor data respectively. The architecture is Data Quality

DQ) relative to accuracy, availability and latency. But, no tangible

mplementation was implemented.

.3. Sensor cloud virtualization

The recent research trends in WSNs aim at expanding the

oundary of using them at large-scale, and integrating them in the

aily human life by making objects surrounding us more proactive

ctors and able to interact and provide services. However, there

re several handicaps that make WSNs not publically exposed and

vailable to end-users at large-scale in particular: (1) deploying,

anaging and maintaining WSNs is usually complex and expen-

ive, and requires a lot of detailed knowledge, (2) WSNs are tradi-

ionally designed to serve one application for a specific purpose.

With the advances in virtualization technologies, there is an

merging vision for using cloud computing to (1) promote the

haring of several physical sensors among multiple users seam-

essly through common interfaces and, (2) provide abstraction lay-

rs on top of physical sensor devices to allow for their access. Sen-

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 269

Fig. 3. Cloud-based sensor monitoring platform [99] .

Fig. 4. Virtual sensor configuration as proposed in [103] .

s

p

t

r

p

m

c

c

o

f

s

f

w

a

c

o

e

v

c

t

p

e

F

a

i

f

t

a

w

a

e

r

fi

p

h

e

p

t

a

S

n

a

c

m

v

E

i

s

c

t

w

t

or clouds help in virtualizing physical sensors on a cloud com-

uting infrastructure and exposing their sensing services via vir-

ual sensors. Doing so, end-users would take benefit of a variety of

emote sensing services without worrying about the locations of

hysical sensors or about their technical details such as manage-

ent or maintenance as it is managed by the cloud.

SOA and Web Services played an important role in this pro-

ess. In [102] , the authors proposed an extensible cloud-based ar-

hitecture for WSNs that uses Open.Sen.se platform, which is an

pen platform that provides APIs to (1) collect, analyze, store dif-

erent sensor data streams using HTTP over IP, and (2) display and

hare collected sensor data using REST-based Web Services inter-

aces, thus enabling interoperable data access anytime from any-

here. The researchers validated their work by collecting temper-

ture and battery voltage readings. They proved that collected data

an be accessed anywhere from any mobile device within 11 sec-

nds on average for the alert email notification and an efficient

nergy consumption on sensor nodes.

Researchers in [103] addressed the problem of physical sensor

irtualization. They defined virtual sensor networks as a “sensor

loud that decouples the network owner and the user and allows mul-

iple WSNs to interoperate at the same time for a single or multi-

le applications that are transparent to the user ”. They have consid-

red four different configurations of virtual sensors as presented in

ig. 4 , and expressed as one-to-many, many-to-one, many-to-many,

nd derived configurations. It captures the cardinality of one phys-

cal sensor with respect to virtual sensors.

The authors also proposed a layered sensor-cloud architecture

or the University of Missouri divided into three main layers: (1)

he client-centric layer, which provides user interfaces and acts as

gateway between the user and the sensor cloud, (2) the middle-

are layer, which mediates between the users and physical sensors

nd performs the virtual sensors management and provisioning to

nd-users and clients, finally, (3) the sensor-centric layer, which di-

ectly interacts with the physical sensors.

The above work presents an interesting contribution to the

eld. However, they have used the old RMI technology for remote

rocedure calls, while the use of REST or SOAP Web Services would

ave been more appropriate for better interoperability and service

xposure.

The Advanced Multi-risk Management SIGMA project [104] was

roposed to manage heterogeneous data collected from various

ypes of sensors and to provide services to prevent and man-

ge risk situations. The authors used cloud computing to enable

ensor and Actuator as a Service (SAaaS), which consists of dy-

amically provisioning sensing resources (device-driven solution)

nd sensed data (data-driven solution) as services on demand ac-

ording to users’ requirements. SAaaS architecture comprises three

ain modules: (1) the Hypervisor which provides abstraction and

irtualization of sensing and actuation resources, (2) the Automatic

nforcer and (3) the Volunteer Cloud , which are responsible for the

nteraction among nodes. The authors have demonstrated the fea-

ibility of their architecture by implementing a monitoring energy

onsumption application in an industrial environment and a smart

raffic control.

In [105] , researchers describe sensor cloud service environment

hich provides customers with remote services provided by a vir-

ual sensor network. They addressed security aspects of the service

270 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

Table 2

A summary of sensor clouds architectures.

Work Scalability Reliability Cost Usability Security

ANGELS [87] yes – yes yes

Integrating sensors with the cloud using dynamic proxies [88] yes no yes – no

Mobile healthcare systems with multi-cloud offloading [90] yes – – yes no

Energy efficient GPS sensing with cloud offloading [91] yes yes – yes no

Healing on the cloud: secure cloud architecture for medical wireless sensor networks [95] yes – – yes yes

Web services framework for wireless sensor networks [98] yes – yes yes yes

Towards a quality-centric big data architecture for federated sensor services [101] yes – – no no

Sensor-cloud: a hybrid framework for remote patient monitoring [96] yes yes – yes yes

A cloud computing solution for patients data collection in health care institutions [97] yes – – yes no

Integrating wireless sensor network into cloud services for real-time data collection [102] yes yes yes yes yes

Sensor cloud: a cloud of virtual sensors [103] yes – – yes yes

SensorCloud: an integrated system for advanced multi-risk management [104] yes yes – yes no

EPIKOUROS platform [106] yes – – yes no

Fig. 5. VANET’s communication modes.

n

t

T

a

u

s

M

o

c

n

v

V

(

6

t

a

a

i

c

T

t

s

v

v

m

s

s

c

delivery from three perspectives, including secure pre-deployment,

secure pre-processing, and secure runtime. EPIKOUROS [106] is a

virtualized platform aiming at integrating multiple heterogeneous

sensor services in a cloud computing environment. The objective

of the platform is to provide new methodologies and tools to de-

sign, build and deploy IoT applications on virtual environments and

manage the business process procedures supported by the sen-

sors’ infrastructure. The platform is based on the publish-subscribe

model to collect sensor measurements. Following a SOA approach,

the platform offers the required services to manipulate/manage

collected information. It allows developers to choose its applica-

tion deployment and the required sensors through a graphical user

interface.

All the above works, among others, contributed to providing an

insight on how to virtualize sensor networks and the benefits un-

derlying such virtualization. However, all these effort s are still in

its early phases and there remains plenty of room for more con-

tributions and research to produce widely accepted solutions. In

particular, there is a crucial need for standards to ensure interop-

erability at public and large-scale.

In Table 2 , the sensor clouds architectures are compared with

respect to the key requirements presented and discussed in

Section 3 .

6. Vehicular cloud

Despite many safety advancements such as airbags, cars remain

more dangerous than other means of transportation. In the United

States, between 20 0 0 and 20 09, accidents involving private users

of cars and light trucks were the cause of 70% of the fatalities

in transportation incidents [107] . Worldwide, road traffic accidents

are the number one cause of death among those of age 15–29

[108] . Vehicular Ad-hoc Networks (VANETs) are being introduced

to enable Cooperative Intelligent Transport Systems (C-ITS) that in-

clude not only information and comfort functions (such as weather

and traffic conditions) but also safety critical applications, such as

vehicle platooning, forward collision warning, and blind intersec-

tion warning systems. A VANET [109,110] is a self-organized net-

work that can be formed by connecting vehicles equipped with

Dedicated Short Range Communications (DSRC). As illustrated in

Fig. 5 , the communication can be between vehicles (V2V) and be-

tween vehicles and infrastructure (V2I), through Road-Side Units

(RSUs).

In Europe, several car manufacturers signed a memorandum re-

garding the intention to make vehicles available with V2x tech-

nologies by 2015 [111] , and, in the USA, the National Highway Traf-

fic Safety Administration (NHSTA) expressed the intention to make

connected vehicle technologies mandatory in light vehicles [112] .

This large-scale deployment of VANETs brings important opportu-

ities to improve safety and comfort functions, and opens the door

o new opportunities by merging cloud computing with VANETs.

he Vehicular Cloud [113] is a new hybrid technology, which takes

dvantage of cloud computing to serve the drivers of VANETs by

sing resources, such as computing, storage and Internet for deci-

ion making.

The term Vehicular Cloud was firstly coined in [114] where

ohamed Eltoweissy et al. defined vehicular cloud as a group

f largely autonomous vehicles whose corporate computing, sensing,

ommunication and physical resources can be coordinated and dy-

amically allocated to authorized users .

In this section, we will review existing contributions to the de-

elopment of the Vehicular Cloud (VC) concept by presenting: (1)

C Design Issues and VC Architecture, (2) Offloading Computation,

3) Big Data, and (4) Virtualization.

.1. VC design issues and VC architectures

Both cloud computing and Vehicular Cloud have similarities in

hat they offer a pool of resources like storage, communication

nd computing. However, Vehicular Clouds have location-relevant

pplications, where the data is originated from the vehicles and

ts surrounding environment. Vehicular Cloud applications are also

haracterized by their mobility, as vehicles are constantly moving.

herefore, pooling of resources must cope with these dynamics.

Some research works have discussed Vehicular Cloud architec-

ures and their requirements. For instance, authors in [113] have

ummarized a typical VC architectures into three layers: inside-

ehicle, communication, and the cloud (see Fig. 6 ). The inside-

ehicle part is responsible for monitoring the driver’s health and

ood and collecting information such as temperature and pres-

ure. The cloud can be used for storing information collected from

ensors and performing complex computations. They have also dis-

ussed some application scenarios and how the cloud will serve

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 271

Fig. 6. Vehicular cloud computing architecture [113] .

s

a

w

a

c

s

p

a

C

r

c

t

t

t

S

f

c

a

s

t

m

a

m

p

[

s

c

m

s

p

a

6

c

a

fi

t

t

t

(

t

v

t

m

v

f

c

t

l

t

D

t

s

V

R

a

c

o

o

s

l

b

t

g

T

i

i

d

c

R

t

o

g

ome VANETs applications in terms of resource allocation, remote

ccess, storage capabilities, and security and privacy.

The work reported in [115] examines the evolution of VANETs

ith two emerging paradigms: Information-Centric Networking

nd Vehicular Cloud Computing. This work defines vehicular cloud

omputing as a set of resources, remotely accessed by vehicles to

hare their data and collaborate to produce some services. They

ropose to leverage Information Centric Networking (ICN) to prop-

gate cloud contents between vehicles. This creates a Vehicular

loud Network (VCN) to offer higher-level vehicular services: cloud

esource discovery, cloud formation, task assignment and results

ollection, and content publishing and sharing.

The DARWIN architecture [116] builds on Service Oriented Archi-

ecture to design a next-generation automotive software platform

hat interacts with cloud services. The architecture is composed of

wo key service components: the Service Process Manager and the

ervice Space, interacting both inside and outside of the vehicle to

orm a vehicular cloud. From design perspective, the authors dis-

ussed three main challenges to be considered in the design of the

rchitecture, namely, (1) real-time guarantees, (2) safety assurance

ecurity, and (3) software engineering issues. The DARWIN archi-

ecture has the benefit of being designed upon conventional auto-

otive components (AUTOSAR) and standard messaging protocols,

nd follows a SOA approach to access and monitor services of auto-

otive devices. It was demonstrated that REST web services reduce

rocessing time up to 40% as compared to SOAP web services.

Another vehicle cloud computing architecture was proposed in

117] , which includes a device level, a communication level and a

ervice level. The architecture focuses on providing real time vehi-

le cloud services such as road traffic monitoring, and health car

onitoring. The main merit of this work is to address real-time

ervices in vehicular networks. However, the weakness of this pa-

er is that it is limited to the conceptual high-level design without

validation or concrete implementation of the architecture.

.2. Offloading computation

Modern cars are well-equipped with a multitude of on-board

omputing devices that enable them to take advantage of cooper-

tive functions that exploit VANETs, such as determining the traf-

c state, and the distance between neighboring vehicles. However,

here is a great potential to improve and enable new functionali-

ies by offloading computation intensive blocks from the vehicles

o the cloud. Computations offloading in VANETs are of two kinds:

1) “between vehicles”: a vehicle may send a part of an application

o another more powerful car to be executed, and (2) between a

ehicle and the infrastructure.

In the last few years, several research works discussed compu-

ation offloading in vehicles. Authors in [118] proposed and imple-

ented a framework for offloading computation for interconnected

ehicles. Vehicles in the proposed framework may select between

our selection strategies, which were based on (1) the computation

apability of each available surrogate vehicle, and (2) the longest

ime interval between inter-accessible vehicle pairs. The four se-

ection strategies proposed in this paper are (1) Random selec-

ion strategy, (2) Computing capacity-based selection strategy, (3)

istance-based selection strategy, and (4) Multi-attributed selec-

ion strategy. Through simulation, the authors have shown that

trategy (4) performs better. This work was targeting Vehicle-to-

ehicle offloading which can be useful in the case of absence of an

SU nearby.

Floating Car Data (FCD) is a method used to estimate the over-

ll traffic conditions based on data (such as position of the vehi-

le, direction, and speed) generated by the cars and mobile phones

f the users in the cars. Authors in [119] have presented the gain

btained by offloading FCD to a cloud environment in order to re-

olve the limitations when managing FCD data by a traditional cel-

ular network paradigm. They have designed a Vehicle-to-Vehicle

ased FCD offloading model in which data is collect by a DSRC and

ransmitted to a set of selected vehicles. In turn, these vehicles

ather the data they receive to upload it via a cellular network.

his work supports the advantages of employing cloud offloading

n such scenario. While the widespread of FCD-based services may

ncrease the uplink load on the cellular networks, the authors have

emonstrated that this approach may decrease traffic load and the

ost of uploading data through local aggregation. Interested in FCD,

azvan Stanica et al. [120] have studied FCD offloading compu-

ation through vehicle-to-vehicle (V2V) communication. The main

bjective of this research was to identify a set of vehicles that

ather and collect FCD data from their neighboring vehicles and

272 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

[

u

f

m

c

s

a

i

t

t

D

t

i

d

6

m

e

a

h

t

[

s

a

m

w

s

a

V

t

m

V

V

i

T

t

t

o

c

i

[

a

c

a

t

w

d

v

w

s

i

p

T

t

T

m

w

p

perform data fusion and upload it to the cloud. Through experi-

ments, authors have proven that this approach achieved good per-

formance.

The work presented in [121] is concerned with developing dis-

aster management system for transportation systems called Intelli-

gent Cloud-based Disaster Management System (ICDMS). The sys-

tem exploit V2X communications to help vehicles transform data

collected from various locations into decisions and strategies. The

proposed system has proven its effectiveness when compared with

a disaster response system. However, these results are verified un-

der very specific conditions (rate transmission, message size, etc),

which cannot prove the effectiveness of the system in real scenar-

ios.

6.3. Big data processing

Today, with the revolution of communication technology, vehi-

cles can communicate, collaborate, and connect with each other

without human’s mediation. This cooperation may generate a huge

volume of information when cars are sending their location, speed,

and environment-related information. Catastrophic accidents have

encouraged vehicle designers to analyze this increasing data to

predict traffic and ensure drivers safety. Big data for vehicles is

about turning the ever-increasing collection of data from various

agents into actionable information [122] . The challenge of this

technology remains in collecting, transmitting, and analyzing this

information and subsequently sending the results to all relevant

target vehicles, all at high speed.

Lionel Nkenyereye et al. [123] presented a study of big data us-

ing hadoop framework to process diagnostic data of a set of con-

nected vehicles. Diagnostic data is collected from android software,

stored in datacenters, and processed by Hadoop. The outcomes are

stored on web servers allowing third party access via web services.

The system presented is a good example of how Big Data process-

ing can be applied in the context of vehicular applications.

A novel multilayered vehicular data cloud platform was pro-

posed recently [124] . The work introduces two vehicular data cloud

services: an intelligent parking cloud service and a vehicular data

mining cloud service. They present two different data mining tech-

niques for the vehicular data mining cloud service, and experimen-

tally show the feasibility of their vehicular data mining service.

This is another interesting Big Data application, but unfortunately

it did not present experiments to evaluate the quality of the ser-

vices.

Yisheng Lv et al. describe in [125] a deep-learning method to

predict traffic flow. Served from a rich amount of traffic flow re-

lated data, they have designed an architecture to extract traffic

flow features. The architecture contains a logistic regression layer

for traffic flow prediction. The accuracy and the effectiveness of the

proposed method are well studied throw experiments and com-

parative study with previous works. Interested in vehicle naviga-

tion information big data analysis, authors in [126] have designed

a model for providing the drivers the appropriate service. The pro-

posed model is mainly composed of (1) a vehicle data collection

module used to collect and transmit data to the collection center

along with the user’s confirmation and, (2) the vehicle informa-

tion center used to manage this data and offer driving informa-

tion. One key advantage of the proposed model is the use of stan-

dard components like OBD-II. Unfortunately, the use of this stan-

dard is limited to a specific list of cars. Moreover, security con-

cerns related to mobile devices should be considered. Authors in

[127] have designed a Storage-as-a-Service system. Vehicles send

data to a database created in roadside units. This data can be

used later by other drivers or vehicle producers. The work present

a parking mall scenario. But, no architectural view and compo-

nents of the system were described. Shengcheng Yuan et al. in

128] were interested in analyzing large-scale traffic evacuations

sing big data techniques to provide better information support

or emergency decision. They have designed a cross-simulation

ethod that, when applied to a huge amount of collected data,

an generate useful information for emergency decision. In a first

tep, this method collects the dynamic distance-speed relation for

ll the drivers. These data are then stored in a database. After cal-

bration, the data distribution is analyzed to determine whether

hey would create an emergency situation. Finally, in the adapta-

ion phase the system can estimate recent driving behavior states.

espite the idea of this research paper is important, it does not

ake into consideration the traffic and road conditions which may

nfluence the acquisition phase.

We note that the systems presented were not designed to pro-

uce real-time answers that can be acted upon.

.4. Virtualization

Virtualization has been used in networking for resource opti-

ization, consolidation, maximizing uptime, workload migration,

tc. But, recently the idea of virtual vehicle was introduced in the

utomotive world. This will enable the electrical system of the ve-

icle to integrate more functions with fewer control devices.

Virtualization already has its place in standard vehicular archi-

ectures such as AUTOSAR (AUTomotive Open System ARchitecture)

129] , which enables the use of a component based software de-

ign model for the design of a vehicular system that can be re-

lized through virtualization that isolates the different functional

odules. AUTOSAR is a good solution to integrate components

ithin an Engine Control Unit (ECU). However, the integration of

oftware inside and outside the vehicles over different networks is

challenging issue.

Authors in [130] proposed a virtualization layer called the

aNetLayer, which defines procedures that enable mobile nodes

o collaborate and create an infrastructure of virtual nodes to

ake the VANET a more reliable communication environment. The

aNetLayer introduced some modifications and additions to the

irtual Node Layer (VNLayer) proposed in [131] to cope with the

nefficiency of the VNLayer constructs when tested with VANETs.

hrough simulations, the authors demonstrated the advantages of

heir work, in that it ensured good packet delivery ratios and bet-

er download time compared to torrent-like solutions.

Innovation in automotive industry has increased the number

f electronic control units (ECU) used in cars. This has led to

omplexity problems in terms of energy consumption, cost, and

nstallation space. To solve this problem, Simon Gansel et al. in

132] proposed to consolidate automotive systems through virtu-

lization to share hardware between different components espe-

ially, graphics hardware. They have designed an architecture of

Virtualized Automotive Graphics System (VAGS) to isolate be-

ween graphics applications running in dedicated VMs. This work

as based on a technical requirement that this architecture was

erived from relevant ISO standards, which is a strong point in fa-

or of its industrial adoption. However, the proposed architecture

as implemented but no performance evaluation or comparative

tudies were presented.

Considering the importance of a vehicular gateway in linking

n-vehicle networks with the Internet, Zonghua Gu et al. [133] pro-

osed an automotive telematics gateway based on virtualization.

he gateway supports two guest operating systems to separate be-

ween the executions of real-time and non real time applications.

he architecture of the gateway performs real-time scheduling to

eet the task management dynamicity. While several applications

ere developed on the gateway, no performance evaluation of the

roposed gateway was presented.

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 273

Table 3

A summary of vehicular cloud architectures.

Work reference Scalability Reliability Cost Usability Security

DARWIN architecture [116] yes yes – no no

Real time services for future cloud computing enabled vehicle networks [117] yes – – yes yes

ICMDS: an intelligent cloud based disaster management system for vehicular networks [121] – yes – yes no

A study of big data solution using hadoop to process connected vehicle’s diagnostics data [123] yes yes yes yes yes

Developing vehicular data cloud services in the IoT environment [124] yes no yes – yes

VaNetLayer: a virtualization layer supporting access to web contents from within vehicular networks [130] yes no – yes no

Towards virtualization concepts for novel automotive HMI systems [132] – – yes yes no

Design and implementation of an automotive telematics gateway based on virtualization [133] – yes – yes no

c

p

7

c

l

w

y

c

a

b

A

“

c

f

s

R

In Table 3 , the architectures for integrating vehicles with cloud

omputing are compared with respect to the key requirements

resented and discussed in Section 3 .

. Discussions and future challenges

This paper presented an overview on the potential use of cloud

omputing to promote cyber-physical applications, and particu-

arly focused on three major areas of applications, namely robotics,

ireless sensor networks and vehicular networks.

Certainly, huge efforts have been carried out over the last five

ears to provide concrete examples on what a cyber-physical cloud

an be. However, the path is still long for this technology to reach

certain maturity as there are several new challenges that must

e addressed. Next, we summarize these challenges:

• Need for standards : For any system to be widely adopted, there

is a crucial requirement to adopt standards to promote interop-

erability among heterogeneous systems and provide universal

solutions that are vendor-independent and platform-agnostic.

The current status of cyber-physical clouds clearly shows that

most of the solutions are proprietary although some of them

rely on open-source platforms, mainly Apache Hadoop with its

Map-Reduce framework for data storage and processing. But

this is not sufficient if universal agreements on networking pro-

tocols, data exchange formats, interfaces, etc, are not fullfilled.

In other words, it is important to unify for each technology a

set of standard protocols that govern the interaction between

cyber-physical systems, Machine-to-Machine and also their in-

teraction with clients and end-users, as well as the storage and

processing processes, and network interfaces. The use of Web

Services, whether REST or SOAP, seems to be very promising

considering their widespread use and their efficiency in ensur-

ing the integration of heterogeneous systems. Their use with

cyber-physical systems can be easily extended to ensure the

cyber-physical system to cloud integration.

• Privacy and security : There has been very little work on cyber-

physical clouds security and privacy issues. Although several

challenges would be shared with traditional cloud computing

systems, cyber-physical clouds would be subject to additional

threats that must be carefully addressed. In fact, considering

the fact that cyber-physical clouds are highly distributed sys-

tems involving different and greatly hereterogenous compo-

nents, attacks can occur at different layers, either at the cyber-

physical layer or the cloud layer. Therefore, end-to-end secu-

rity mechanisms must be conceived and implemented to en-

sure the integrity of any transaction that takes place in the

cyber-physical cloud. This is rather a crucial requirement as any

vulnerability may have serious consequences on safety-critical

applications. For example, an attacker may take over a phys-

ical sensor infrastructure and generate faulty data, then dis-

rupt the system operation, or alternatively gain access to the

cloud infrastructure and get access to unauthorized data. Secu-

rity problems in clouds for sensor services were addressed in

recent work [134] and also discussed earlier in [135] . On the

other hand, privacy represents a major concern as data emanat-

ing from private cyber-physical system sources is stored in the

cloud, which can be public or private, and presents the thread

to be access by unauthorized tiers.

• Real-time : Cyber-physical clouds encompass time-sensitive ap-

plications that require real-time guarantees to deliver time-

critical data, in particular for automation and control applica-

tions. For example, in mission-critical scenarios where sensor

nodes populate the cloud with their data to take appropriate

actions to be executed by actuators or robots, there is a need to

have a bounded end-to-end delay to guarantee the correctness

of the system operation. The concern becomes even more seri-

ous when data grows at large-scale. There is a need to carefully

perform the dimensioning of the system in advance to ensure

the quality of service needed for the operation of time-critical

applications. Resource provisioning, scheduling and differenti-

ated services are examples of possible research axes in this re-

gard.

• Programming abstractions : Considering the big data scale in

cyber-physical clouds, current programming abstractions might

be insufficient to sufficiently scale with the size of the data to

be processed. Although Map-Reduce framework represents a

defacto standard for processing large data, there is still needs

for companion mechanisms to filter out data coming from het-

erogeneous sources and thus reducing the computation space

complexity through effective sampling without compromising

the outcomes. In addition, it is important to provide developers

with new Application Programming Interfaces (APIs) that allow

them to easily interact with cyber-physical clouds, in the same

way as public clouds, like Amazon Web Services, Google Apps

Engine, Facebook and others, are providing for developers.

There are currently some efforts and solutions in this respect

like the Sensor API and Samsung GALAXY SDK for Android

developers.

cknowledgement

This work is supported by the Dronemap project entitled

DroneMap: A Cloud Robotics System for Unmanned Aerial Vehi-

les in Surveillance Applications” under the grant number 35–157

rom King Abdul Aziz City for Science and Technology (KACST), and

upported by Prince Sultan University.

eferences

[1] J.K. Muppala, M. Hiltunen, R. Stroud, J. Wang, The first international work-shop on dependability of clouds, data centers and virtual computing environ-

ments, in: Dependable Systems Networks (DSN), 2011 IEEE/IFIP 41st Interna-

tional Conference on, 2011, pp. 590–591, doi: 10.1109/DSN.2011.5958232 . [2] M.B. Mollah, K.R. Islam, S.S. Islam, Next generation of computing through

cloud computing technology, in: Electrical Computer Engineering (CCECE),2012 25th IEEE Canadian Conference on, 2012, pp. 1–6, doi: 10.1109/CCECE.

2012.6334973 . [3] C. Ji, Y. Li, W. Qiu, U. Awada, K. Li, Big data processing in cloud computing

environments, in: Pervasive Systems, Algorithms and Networks (ISPAN), 2012

12th International Symposium on, 2012, pp. 17–23, doi: 10.1109/I-SPAN.2012.9 .

274 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

[4] Q. Duan, Y. Yan, A.V. Vasilakos, A survey on service-oriented network virtual-ization toward convergence of networking and cloud computing, IEEE Trans.

Netw. Serv. Manage. 9 (4) (2012) 373–392, doi: 10.1109/TNSM.2012.113012.120310 .

[5] R. Rajkumar , I. Lee , L. Sha , J. Stankovic , Cyber-physical systems: the nextcomputing revolution, in: Design Automation Conference (DAC), 2010 47th

ACM/IEEE, 2010, pp. 731–736 . [6] B. Qureshi, A. Kouba, Five traits of performance enhancement using cloud

robotics: a survey, Procedia Comput. Sci. 37 (2014) 220–227 . The 5th In-

ternational Conference on Emerging Ubiquitous Systems and Pervasive Net-works (EUSPN-2014)/ The 4th International Conference on Current and Fu-

ture Trends of Information and Communication Technologies in Health-care (ICTH 2014)/ Affiliated Workshops, URL http://www.sciencedirect.com/

science/article/pii/S1877050914009983 . http://dx.doi.org/10.1016/j.procs.2014.08.033 .

[7] B. Kehoe, S. Patil, P. Abbeel, K. Goldberg, A survey of research on cloud

robotics and automation, IEEE Trans. Autom. Sci. Eng. 12 (2) (2015) 398–409,doi: 10.1109/TASE.2014.2376492 .

[8] R. Doriya, S. Mishra, S. Gupta, A brief survey and analysis of multi-robotcommunication and coordination, in: Computing, Communication Automation

(ICCCA), 2015 International Conference on, 2015, pp. 1014–1021, doi: 10.1109/CCAA.2015.7148524 .

[9] R.C. Chandrani , A survey on cloud sensor integration, Int. J. Innov. Res. Com-

put. Commun. Eng. (IJIRCCE) 2 (August) (2014) 5470–5476 . [10] K. Sindhanaiselvan , T. Mekala , A survey on sensor cloud: architecture and ap-

plications, Int. J. P2P Netw. Trends Technol. (IJPTT) 6 (March) (2014) 49–53 . [11] I. Khan, F. Belqasmi, R. Glitho, N. Crespi, M. Morrow, P. Polakos, Wireless

sensor network virtualization: a survey, IEEE Commun. Surv. Tutorials 18 (1)(2016) 553–576, doi: 10.1109/COMST.2015.2412971 .

[12] M. Abuelela, S. Olariu, Taking vanet to the clouds, in: Proceedings of the 8th

International Conference on Advances in Mobile Computing and Multimedia,in: MoMM ’10, ACM, New York, NY, USA, 2010, pp. 6–13, doi: 10.1145/1971519.

1971522 . [13] G. Mario , Vehicular cloud computing, in: Vehicular Communications and Ap-

plications Workshop, IEEE, 2012, pp. 152–155 . [14] J. Hu, A. Edsinger, Y.-J. Lim, N. Donaldson, M. Solano, A. Solochek, R. Marches-

sault, An advanced medical robotic system augmenting healthcare capabil-

ities - robotic nursing assistant, in: Robotics and Automation (ICRA), 2011IEEE International Conference on, 2011, pp. 6264–6269, doi: 10.1109/ICRA.2011.

5980213 . [15] Y. Khaled, M. Tsukada, J. Santa, T. Ernst, On the design of efficient vehicu-

lar applications, in: Vehicular Technology Conference, 2009. VTC Spring 2009.IEEE 69th, 2009, pp. 1–5, doi: 10.1109/VETECS.2009.5073727 .

[16] K. Supun, C. Leif, F. Geoffrey, A framework for real time processing of sensor

data in the cloud, J.f Sens. 2015 (2015) 11, doi: 10.1155/2015/468047 . [17] H. Hromic, D.L. Phuoc, M. Serrano, A. Antoni, I.P. arko, C. Hayes, S. Decker,

Real time analysis of sensor data for the internet of things by means of clus-tering and event processing, in: 2015 IEEE International Conference on Com-

munications (ICC), 2015, pp. 685–691, doi: 10.1109/ICC.2015.7248401 . [18] O. Mutlu, L. Subramanian, Research problems and opportunities in mem-

ory systems, Supercomput. Front. Innov. 1 (3) (2015) . URL http://superfri.org/superfri/article/view/30

[19] O. Mutlu , J. Meja , L. Subramanian , The main memory system: challenges and

opportunities, Invited Article in Communications of the Korean Institute ofInformation Scientists and Engineers (KIISE) (2015) 16–41 .

[20] L. Banda, M. Mzyece, G. Noel, Ip mobility support: solutions for vehicularnetworks, IEEE Veh. Technol. Mag. 7 (4) (2012) 77–87, doi: 10.1109/MVT.2012.

2203881 . [21] H. Fotouhi, M. Alves, M.Z. Zamalloa, A. Kouba, Reliable and fast hand-offs in

low-power wireless networks, IEEE Trans. Mob. Comput. 13 (11) (2014) 2620–

2633, doi: 10.1109/TMC.2014.2307867 . [22] N. Baccour, A. Koubâa, L. Mottola, M.A. Zúñiga, H. Youssef, C.A. Boano,

M. Alves, Radio link quality estimation in wireless sensor networks: a survey,ACM Trans. Sen. Netw. 8 (4) (2012) 34:1–34:33, doi: 10.1145/2240116.2240123 .

[23] B. Kantarci, H.T. Mouftah, Mobility-aware trustworthy crowdsourcing incloud-centric internet of things, in: 2014 IEEE Symposium on Computers and

Communications (ISCC), 2014, pp. 1–6, doi: 10.1109/ISCC.2014.6912581 .

[24] The national institute of standards and technology (nist), ( http://www.nist.gov/ ).

[25] V. Kishore , G. Bhavani , M. Ramakrishna , Simplifying complex tasks: cloudcomputing, (IJCSIT) Int. J. Comput. Sci. Inf. Technol. 3 (2012) 4 988–4 996 .

[26] The amazon web services, ( http://aws.amazon.com/ ). [27] Vmware, ( http://www.vmware.com/ ).

[28] Google cloud platform, ( https://cloud.google.com ).

[29] Apps.gov: Federal cloud computing information, ( http://cloud.cio.gov/ ). [30] A. Ghobadi, R. Karimi, F. Heidari, M. Samadi, Cloud computing, reliability

and security issue, in: Advanced Communication Technology (ICACT), 201416th International Conference on, 2014, pp. 504–511, doi: 10.1109/ICACT.2014.

6779012 . [31] D. Zissis, D. Lekkas, Addressing cloud computing security issues, Fu-

ture Gener. Comput. Syst. 28 (3) (2012) 583–592. http://dx.doi.org/10.

1016/j.future.2010.12.006 . URL http://www.sciencedirect.com/science/article/pii/S0167739X10 0 02554 .

[32] L.M. Kaufman, Data security in the world of cloud computing, IEEE Secur. Pri-vacy 7 (4) (2009) 61–64, doi: 10.1109/MSP.2009.87 .

[33] G. Yan, D. Wen, S. Olariu, M.C. Weigle, Security challenges in vehicular cloud

computing, IEEE Trans. Intell. Transp. Syst. 14 (1) (2013) 284–294, doi: 10.1109/TITS.2012.2211870 .

[34] Z. Xiao, Y. Xiao, Security and privacy in cloud computing, IEEE Commun. Surv.Tutorials 15 (2) (2013) 843–859, doi: 10.1109/SURV.2012.060912.00182 .

[35] N. Vrek , S. Brumec , Role of utilization rate on cloud computing cost effectiv-ness analysis, in: Information Society (i-Society), 2013 International Confer-

ence on, 2013, pp. 177–181 . [36] S.-H. Baeg, J.-H. Park, J. Koh, K.-W. Park, M.-H. Baeg, Building a smart home

environment for service robots based on rfid and sensor networks, in: Con-

trol, Automation and Systems, 2007. ICCAS ’07. International Conference on,2007, pp. 1078–1082, doi: 10.1109/ICCAS.2007.4407059 .

[37] C. Francesco , M. Fulvio , S. Antonio , Z. Renato , The anser project: airport non-stop surveillance expert robot, in: Intelligent Robots and Systems, IROS 2007,

IEEE/RSJ International Conference on, 2007 . [38] M. Shiomi, T. Kanda, D.F. Glas, S. Satake, H. Ishiguro, N. Hagita, Field trial

of networked social robots in a shopping mall, in: Intelligent Robots and Sys-

tems, 20 09. IROS 20 09. IEEE/RSJ International Conference on, 20 09, pp. 2846–2853, doi: 10.1109/IROS.2009.5354242 .

[39] The international federation of robotics, ( http://www.ifr.org/service-robots/statistics/ ).

[40] D.M.P. Matei, The study of cooperative robots behaviors in automatic indus-trial manufacturing, in: Applied and Theoretical Electricity (ICATE), 2012 In-

ternational Conference on, 2012, pp. 1–6, doi: 10.1109/ICATE.2012.6403470 .

[41] H. Baalbaki, M. Lamiri, X. Xie, Joint location and configuration of mobileservice robots for hospitals, in: Automation Science and Engineering, 2008.

CASE 2008. IEEE International Conference on, 2008, pp. 615–620, doi: 10.1109/COASE.2008.4626452 .

[42] K. Goldberg, B. Chen, R. Solomon, S. Bui, B. Farzin, J. Heitler, D. Poon, G. Smith,Collaborative teleoperation via the internet, in: Robotics and Automation,

20 0 0. Proceedings. ICRA ’00. IEEE International Conference on, vol. 2, 20 0 0,

pp. 2019–2024, doi: 10.1109/ROBOT.20 0 0.844891 . [43] K. Goldberg, M. Mascha, S. Gentner, N. Rothenberg, C. Sutter, J. Wiegley, Desk-

top teleoperation via the world wide web, in: Robotics and Automation, 1995.Proceedings., 1995 IEEE International Conference on, vol. 1, 1995, pp. 654–

659, doi: 10.1109/ROBOT.1995.525358 . [44] G. McKee , What is networked robotics? in: Informatics in Control Automation

and Robotics, in: Lecture Notes Electrical Engineering, Springer Berlin Heidel-

berg, 2008, pp. 35–45 . [45] J. Kuffner , Cloud-enabled robots, in: IEEE-RAS International Conference on

Humanoid Robots, IEEE, 2010 . [46] D.P. Donato , M. Annalisa , C. Grazia , D. Arcangelo , An autonomous mobile

robotic system for surveillance of indoor environments, Int. J. Adv. Rob. Syst.7 (2010) 19–26 .

[47] G. Mohanarajah, D. Hunziker, R. D’Andrea, M. Waibel, Rapyuta: a cloud

robotics platform, IEEE Trans. Autom. Sci. Eng. 12 (2) (2015) 4 81–4 93, doi: 10.1109/TASE.2014.2329556 .

[48] L. Bingwei , C. Yu , B. Erik , P. Khanh , S. Dan , C. Genshe , A holistic cloud-enabledrobotics system for real-time video tracking application, in: Future Informa-

tion Technology, in: Lecture Notes in Electrical Engineering, vol. 276, SpringerBerlin Heidelberg, 2014, pp. 455–468 .

[49] T. Shuqing, S. Dilshat, G.L. Suk, Cloud robot with real-time face recognitionability, Adv. Sci. Technol. Lett. 51 (2014) 77–80. http://dx.doi.org/10.14257/astl.

2014.51.18 .

[50] L. Turnbull, B. Samanta, Cloud robotics: formation control of a multi robotsystem utilizing cloud infrastructure, in: Southeastcon, 2013 Proceedings of

IEEE, 2013, pp. 1–4, doi: 10.1109/SECON.2013.6567422 . [51] R. Doriya, P. Chakraborty, G.C. Nandi, Robotic services in cloud computing

paradigm, in: Cloud and Services Computing (ISCOS), 2012 International Sym-posium on, 2012, pp. 80–83, doi: 10.1109/ISCOS.2012.24 .

[52] R. Arumugam, V.R. Enti, L. Bingbing, W. Xiaojun, K. Baskaran, F.F. Kong,

A.S. Kumar, K.D. Meng, G.W. Kit, Davinci: a cloud computing framework forservice robots, in: Robotics and Automation (ICRA), 2010 IEEE International

Conference on, 2010, pp. 3084–3089, doi: 10.1109/ROBOT.2010.5509469 . [53] Y. Nimmagadda, K. Kumar, Y.H. Lu, C.S.G. Lee, Real-time moving object recog-

nition and tracking using computation offloading, in: Intelligent Robots andSystems (IROS), 2010 IEEE/RSJ International Conference on, 2010, pp. 2449–

2455, doi: 10.1109/IROS.2010.5650303 .

[54] V. Ovidiu , F. Peter , Internet of Things: Converging Technologies for Smart En-vironments and Integrated Ecosystems, River Publishers Series in Communi-

cation, 2013 . [55] The robo brain project, ( http://robobrain.me/ ).

[56] The columbia grasp database, ( http://grasping.cs.columbia.edu/ ). [57] G. Corey , C. Matei , D. Hao , K.A. Peter , The columbia grasp database, in: Inter-

national Conference on Robotics and Automation, 2009 .

[58] M. Sato, K. Kamei, S. Nishio, N. Hagita, The ubiquitous network robot plat-form: Common platform for continuous daily robotic services, in: System In-

tegration (SII), 2011 IEEE/SICE International Symposium on, 2011, pp. 318–323, doi: 10.1109/SII.2011.6147467 .

[59] The willow garage, personal service robotics with tiered human-in-the-loopassistance, 2013. ( https://www.sbir.gov/sbirsearch/detail/391629 ).

[60] Google goggles, ( http://www.google.com/mobile/goggles/ ).

[61] Z. Dogmus, E. Erdem, V. Patoglu, Rehabrobo-onto: design, development andmaintenance of a rehabilitation robotics ontology on the cloud, Rob. Com-

put. Integr. Manuf. 33 (2015) 100–109 . Special Issue on Knowledge DrivenRobotics and Manufacturing. http://dx.doi.org/10.1016/j.rcim.2014.08.010 .

URL http://www.sciencedirect.com/science/article/pii/S07365845140 0 0714 .

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 275

[

[

[

[62] The roboearth project, ( http://roboearth.org/ ). [63] M. Waibel, M. Beetz, J. Civera, R. D’Andrea, J. Elfring, D. Galvez-Lopez,

K. Haussermann, R. Janssen, J.M.M. Montiel, A. Perzylo, B. Schiessle,M. Tenorth, O. Zweigle, R. van de Molengraft, Roboearth, Rob. Autom. Mag.

IEEE 18 (2) (2011) 69–82, doi: 10.1109/MRA.2011.941632 . [64] T. Tsuji, N. Kurita, M. Yamada, S. Sakaino, Y. Kaneko, Rehabilitation database

based on haptic signal processing, in: Humanitarian Technology Conference(R10-HTC), 2013 IEEE Region 10, 2013, pp. 13–18, doi: 10.1109/R10-HTC.2013.

6669006 .

[65] K. Ben , M. Akihiro , C. Sal , K. James , G. Ken , Cloud-based robot grasping withthe google object recognition engine, in: Robotics and Automation (ICRA),

2013 IEEE International Conference on, IEEE, 2013 . [66] S. Nakagawa, N. Igarashi, Y. Tsuchiya, M. Narita, Y. Kato, An implementation

of a distributed service framework for cloud-based robot services, in: IECON2012 - 38th Annual Conference on IEEE Industrial Electronics Society, 2012,

pp. 4148–4153, doi: 10.1109/IECON.2012.6389225 .

[67] C. Mouradian, F.Z. Errounda, F. Belqasmi, R. Glitho, An infrastructure forrobotic applications as cloud computing services, in: Internet of Things (WF-

IoT), 2014 IEEE World Forum on, 2014, pp. 377–382, doi: 10.1109/WF-IoT.2014.6803195 .

[68] S. Mahmoud, N. Mohamed, Collaborative uavs cloud, in: Unmanned AircraftSystems (ICUAS), 2014 International Conference on, 2014, pp. 365–373, doi: 10.

1109/ICUAS.2014.6842275 .

[69] S. Mahmoud, N. Mohamed, Broker architecture for collaborative uavs cloudcomputing, in: Collaboration Technologies and Systems (CTS), 2015 Interna-

tional Conference on, 2015, pp. 212–219, doi: 10.1109/CTS.2015.7210423 . [70] A. Koubaa, Architecture of Computing Systems–ARCS 2014: 27th International

Conference, Lübeck, Germany, February 25-28, 2014. Proceedings, Springer In-ternational Publishing, Cham. Ch. A Service-Oriented Architecture for Virtual-

izing Robots in Robot-as-a-Service Clouds, 2014, pp. 196–208 . doi: 10.1007/

978- 3- 319- 04891- s9 _ 17 . [71] S. Osentoski, G. Jay, C. Crick, B. Pitzer, C. DuHadway, O.C. Jenkins, Robots as

web services: reproducible experimentation and application development us-ing rosjs, in: Robotics and Automation (ICRA), 2011 IEEE International Confer-

ence on, 2011, pp. 6078–6083, doi: 10.1109/ICRA.2011.5980464 . [72] Y. Kato, T. Izui, Y. Murakawa, K. Okabayashi, M. Ueki, Y. Tsuchiya, M. Narita,

Research and development environments for robot services and its imple-

mentation, in: System Integration (SII), 2011 IEEE/SICE International Sympo-sium on, 2011, pp. 306–311, doi: 10.1109/SII.2011.6147465 .

[73] Y. Chen, Z. Du, M. Garca-Acosta, Robot as a service in cloud computing, in:Service Oriented System Engineering (SOSE), 2010 Fifth IEEE International

Symposium on, 2010, pp. 151–158, doi: 10.1109/SOSE.2010.44 . [74] P. Rawat, K. Singh, H. Chaouchi, J. Bonnin, Wireless sensor networks: a sur-

vey on recent developments and potential synergies, J. Supercomput. 68 (1)

(2014) 1–48, doi: 10.1007/s11227- 013- 1021- 9 . [75] C.-Y. Chong, S.P. Kumar, Sensor networks: evolution, opportunities, and chal-

lenges, Proc. IEEE 91 (8) (2003) 1247–1256, doi: 10.1109/JPROC.2003.814918 . [76] K. Divi, M.R. Kanjee, H. Liu, Secure architecture for healthcare wireless sensor

networks, in: Information Assurance and Security (IAS), 2010 Sixth Interna-tional Conference on, 2010, pp. 131–136, doi: 10.1109/ISIAS.2010.5604192 .

[77] C. Oliveira, G. Gonalves, Environmental monitoring services and wireless sen-sor networks applied on urban spaces, in: Automation Quality and Testing

Robotics (AQTR), 2010 IEEE International Conference on, vol. 3, 2010, pp. 1–6,

doi: 10.1109/AQTR.2010.5520681 . [78] M.P. urii , Z. Tafa , G. Dimi , V. Milutinovi , A survey of military applications of

wireless sensor networks, in: Embedded Computing (MECO), 2012 Mediter-ranean Conference on, 2012, pp. 196–199 .

[79] A .A .K. Somappa, K. vsthus, L.M. Kristensen, An industrial perspective on wire-less sensor networks 2014; a survey of requirements, protocols, and chal-

lenges, IEEE Commun. Surv. Tutorials 16 (3) (2014a) 1391–1412, doi: 10.1109/

SURV.2014.012114.0 0 058 . [80] A .A .K. Somappa, K. vsthus, L.M. Kristensen, An industrial perspective on wire-

less sensor networks; a survey of requirements, protocols, and challenges,IEEE Commun. Surv. Tutorials 16 (3) (2014b) 1391–1412, doi: 10.1109/SURV.

2014.012114.0 0 058 . [81] R.V.P. Yerra, A.K. Bharathi, P. Rajalakshmi, U.B. Desai, Wsn based power mon-

itoring in smart grids, in: Intelligent Sensors, Sensor Networks and Infor-

mation Processing (ISSNIP), 2011 Seventh International Conference on, 2011,pp. 401–406, doi: 10.1109/ISSNIP.2011.6146589 .

[82] Rpl: Ipv6 routing protocol for low-power and lossy networks, ( https://tools.ietf.org/html/rfc6550 ).

[83] 6lowpan ad hoc on-demand distance vector routing (load), ( https://tools.ietf.org/html/draft- daniel- 6lowpan- load- adhoc- routing- 03 ).

[84] M. Kovatsch, S. Duquennoy, A. Dunkels, A low-power coap for contiki, in: Mo-

bile Adhoc and Sensor Systems (MASS), 2011 IEEE 8th International Confer-ence on, 2011, pp. 855–860, doi: 10.1109/MASS.2011.100 .

[85] S. Raza, H. Shafagh, K. Hewage, R. Hummen, T. Voigt, Lithe: lightweight se-cure coap for the internet of things, Sensors J. IEEE 13 (10) (2013) 3711–3720,

doi: 10.1109/JSEN.2013.2277656 . [86] A . Atif, A . Wasai Shadab, H. Mohammad Mehedi, H. M. Shamim, A. Abdul-

hameed, H. M. Anwar, A survey on sensor-cloud: architecture, applications,

and approaches, Int. J. Distrib. Sens. Netw. 2013 (2012). http://dx.doi.org/10.1155/2013/917923 .

[87] P. Datta, S. Dey, H.S. Paul, A. Mukherjee, Angels: a framework for mobile grids,in: Applications and Innovations in Mobile Computing (AIMoC), 2014, 2014,

pp. 15–20, doi: 10.1109/AIMOC.2014.6785513 .

[88] W. Wang, K. Lee, D. Murray, Integrating sensors with the cloud using dynamicproxies, in: Personal Indoor and Mobile Radio Communications (PIMRC), 2012

IEEE 23rd International Symposium on, 2012, pp. 1466–1471, doi: 10.1109/PIMRC.2012.6362579 .

[89] The eucalyptus private cloud, ( https://www.eucalyptus.com/ ). [90] H. Wu, Q. Wang, K. Wolter, Mobile healthcare systems with multi-cloud of-

floading, in: Mobile Data Management (MDM), 2013 IEEE 14th InternationalConference on, vol. 2, 2013, pp. 188–193, doi: 10.1109/MDM.2013.92 .

[91] J. Liu, B. Priyantha, T. Hart, H.S. Ramos, A .A .F. Loureiro, Q. Wang, Energy ef-

ficient gps sensing with cloud offloading, in: Proceedings of the 10th ACMConference on Embedded Network Sensor Systems, in: SenSys ’12, ACM, New

York, NY, USA, 2012, pp. 85–98, doi: 10.1145/2426656.2426666 . [92] The great sensor-era: Brontobytes will change society, ( https://datafloq.com/ ).

[93] M. Hirafuji, A strategy to create agricultural big data, in: Global Conference(SRII), 2014 Annual SRII, 2014, pp. 249–250, doi: 10.1109/SRII.2014.43 .

[94] K.D. Sanjit, M. Subasish, K.P. Prasant, A survey on applications of wireless sen-

sor network using cloud computing, Int. J. Comput. Sci. Emerging Technol. 1(December) (2010) 50–55. http://dx.doi.org/10.1155/2013/917923 .

[95] L. Ahmed, H. Abdelkrim, B. Abdelmadjid, C. Yacine, Healing on the cloud: se-cure cloud architecture for medical wireless sensor networks, Future Gener.

Comput. Syst. (2015). http://dx.doi.org/10.1016/j.future.2015.01.009 . [96] M. Subasish , R. K. Smruti , Sensor-cloud: a hybrid framework for remote pa-

tient monitoring, Int. J. Comput. Appl. 55 (2012) 7–11 .

[97] C.O. Rolim, F.L. Koch, C.B. Westphall, J. Werner, A. Fracalossi, G.S. Salvador, Acloud computing solution for patient’s data collection in health care institu-

tions, in: eHealth, Telemedicine, and Social Medicine, 2010. ETELEMED ’10.Second International Conference on, 2010, pp. 95–99, doi: 10.1109/eTELEMED.

2010.19 . [98] A.G. Mark , N.S. Philip , Web services framework for wireless sensor networks,

in: Service Computation 2014 : The Sixth International Conferences on Ad-

vanced Service Computing, 2014, pp. 15–23 . [99] E. Vincent C. , F. Kaniz , H. Philip , M. John P. , Towards a generic cloud-based

sensor data management platform: a survey and conceptual architecture, in:The Eighth International Conference on Sensor Technologies and Applications,

2014, pp. 88–96 . 100] A. Cuzzocrea, G. Fortino, O. Rana, Managing data and processes in cloud-

enabled large-scale sensor networks: state-of-the-art and future research di-

rections, in: Cluster, Cloud and Grid Computing (CCGrid), 2013 13th IEEE/ACMInternational Symposium on, 2013, pp. 583–588, doi: 10.1109/CCGrid.2013.116 .

[101] L. Ramaswamy, V. Lawson, S.V. Gogineni, Towards a quality-centric big dataarchitecture for federated sensor services, in: Big Data (BigData Congress),

2013 IEEE International Congress on, 2013, pp. 86–93, doi: 10.1109/BigData.Congress.2013.21 .

[102] R. Piyare, S. Park, S.Y. Maeng, S.H. Park, S.C. Oh, S.G. Choi, H.S. Choi, S.R. Lee,

Integrating wireless sensor network into cloud services for real-time data col-lection, in: ICT Convergence (ICTC), 2013 International Conference on, 2013,

pp. 752–756, doi: 10.1109/ICTC.2013.6675470 . [103] S. Madria, V. Kumar, R. Dalvi, Sensor cloud: a cloud of virtual sensors, IEEE

Software 31 (2) (2014) 70–77, doi: 10.1109/MS.2013.141 . 104] A. Puliafito, Sensorcloud: an integrated system for advanced multi-risk man-

agement, in: Network Cloud Computing and Applications (NCCA), 2014 IEEE3rd Symposium on, 2014, pp. 1–8, doi: 10.1109/NCCA.2014.10 .

[105] P. Nayot , K. Vimal , M. Sanjay , C. Sriram , Challenges in secure sensor-cloud

computing, in: Secure Data Management, in: Lecture Notes in Computer Sci-ence, 2011, pp. 70–84 .

106] V. Demosthenes , M. Athanasios , S. Manolis , D. Dimitris , L. George , K. Sofok-lis , D. Dimitris , Epikouros - virtualized platforms using heterogeneous sensor

services in cloud computing, in: Wireless Communications, Vehicular Tech-nology, Information Theory and Aerospace and Electronic Systems (VITAE),

4th International Conference on, 2014 .

[107] I. Savage, Comparing the fatality risks in united states transportation acrossmodes and over time, Res. Transp. Econ. 43 (1) (2013) 9–22 . The Economics of

Transportation Safety. http://dx.doi.org/10.1016/j.retrec.2012.12.011 . URL http: //www.sciencedirect.com/science/article/pii/S0739885912002156 .

[108] W.H. Organization , et al. , WHO global status report on road safety 2013: sup-porting a decade of action, World Health Organization, 2013 .

[109] S.R. Ram, K. Manish, S. Nanhay, Security challenges, issues and their solutions

for vanet, Int. J. Network Secur. Appl. (IJNSA) 5 (5) (2013) 95–105, doi: 10.5121/ijnsa.2013.5508 .

[110] A. Marwa , M. Imad , A survey of vehicular ad hoc networks routing protocols,Int. J. Innovation Appl. Stud. 3 (2013) 829–846 .

[111] Vehicle manufacturers signing a memorandum of understanding (mou),( https://www.car- 2- car.org/index.php?id=214 ).

[112] S.B. Valerie , W. Richard , C. Joshua , S. Matt , T. William , Expert opinion fore-

cast of connected vehicle technology, Technical Report, Center for AutomotiveResearch, Michigan Department of Transportation, 2013 .

[113] M. Whaiduzzaman, M. Sookhak, A. Gani, R. Buyya, A survey on vehicularcloud computing, J. Netw. Comput. Appl. 40 (2014) 325–344, doi: 10.1016/j.

jnca.2013.08.004 . [114] O. Stephan , E. Mohamed , Y. Mohamed , Towards autonomous vehicular clouds,

in: Ad Hoc Networks, in: Lecture Notes of the Institute for Computer Sciences,

Social Informatics and Telecommunications Engineering, 2010, pp. 1–16 . [115] E. Lee, E.-K. Lee, M. Gerla, S.Y. Oh, Vehicular cloud networking: architecture

and design principles, IEEE Commun. Mag. 52 (2) (2014) 148–155, doi: 10.1109/MCOM.2014.6736756 .

276 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

[116] A. Iwai, M. Aoyama, Automotive cloud service systems based on service-oriented architecture and its evaluation, in: Cloud Computing (CLOUD), 2011

IEEE International Conference on, 2011, pp. 638–645, doi: 10.1109/CLOUD.2011.119 .

[117] J. Wang, J. Cho, S. Lee, T. Ma, Real time services for future cloud comput-ing enabled vehicle networks, in: Wireless Communications and Signal Pro-

cessing (WCSP), 2011 International Conference on, 2011, pp. 1–5, doi: 10.1109/WCSP.2011.6096957 .

[118] L. Bo , P. Yijian , W. Hao , L. Zhi , L. Haixia , Computation offloading management

for vehicular ad hoc cloud, in: Algorithms and Architectures for Parallel Pro-cessing, in: Lecture Notes in Computer Science, 2014, pp. 728–739 .

[119] S. Ancona, R. Stanica, M. Fiore, Performance boundaries of massive float-ing car data offloading, in: Wireless On-demand Network Systems and Ser-

vices (WONS), 2014 11th Annual Conference on, 2014, pp. 89–96, doi: 10.1109/WONS.2014.6814727 .

[120] R. Stanica, M. Fiore, F. Malandrino, Offloading floating car data, in: World

of Wireless, Mobile and Multimedia Networks (WoWMoM), 2013 IEEE 14thInternational Symposium and Workshops on a, 2013, pp. 1–9, doi: 10.1109/

WoWMoM.2013.6583391 . [121] A . Zubaida , B.A . Mohmmad , A . Saleh , V. Anna Maria , M. Rashid , Icdms: an

intelligent cloud based disaster management system for vehicular networks,in: Communication Technologies for Vehicles, in: Lecture Notes in Computer

Science, 2012, pp. 40–56 .

[122] B.G. Gebeyehu , C. Yi , Y. Zhimin , The paradigm of big data for augmentinginternet of vehicle into the intelligent cloud computing systems, in: Internet

of Vehicles Technologies and Services, in: Lecture Notes in Computer Science,2014, pp. 247–261 .

[123] N. Lionel , J. Jong-Wook , A study of big data solution using hadoop to pro-cess connected vehicle’s diagnostics data, in: Information Science and Appli-

cations, Springer Berlin Heidelberg, 2015, pp. 697–704 .

[124] W. He, G. Yan, L.D. Xu, Developing vehicular data cloud services in the iotenvironment, IEEE Trans. Ind. Inf. 10 (2) (2014) 1587–1595, doi: 10.1109/TII.

2014.2299233 .

[125] Y. Lv, Y. Duan, W. Kang, Z. Li, F.Y. Wang, Traffic flow prediction with big data:a deep learning approach, IEEE Trans. Intell. Transp. Syst. 16 (2) (2015) 865–

873, doi: 10.1109/TITS.2014.2345663 . [126] J. Hae-Jong , H. Suck-Joo , P. Dong-Su , Smart-contents visualization of vehicle

big data using vehicle navigation status information, in: Advanced in Com-puter Science and its Applications: CSA 2013, in: Lecture Notes in Electrical

Engineering, 2014, pp. 4 83–4 88 . [127] M. Monica , L. Chaitrali , G. Pramila , N. Uma , A. Ashwini , Implementing storage

as a service in vanet using cloud environment, Int. J. Adv. Found. Res. Comput.

(IJAFRC) 1 (2014) 70–76 . [128] Y. Shengcheng , L. Yi , W. Gangqiao , Z. Hui , A cross-simulation method for

large-scale traffic evacuation with big data, in: Web-Age Information Man-agement, in: Lecture Notes in Computer Science, 2014, pp. 14–21 .

[129] Autosar (automotive open system architecture), ( http://www.autosar.org/ ). [130] B.-T. Jack F., L.-N. Martin, B.-F. Yolanda, J.P.-A. Jose, F.O.-M. Esteban, Vanet-

layer: a virtualization layer supporting access to web contents from within

vehicular networks, J. Comput. Sci. (2014), doi: 10.1016/j.jocs.2014.07.004 . [131] M. Brown, S. Gilbert, N. Lynch, C. Newport, T. Nolte, M. Spindel, The virtual

node layer: a programming abstraction for wireless sensor networks, SIGBEDRev. 4 (3) (2007) 7–12, doi: 10.1145/1317103.1317105 .

[132] G. Simon , S. Stephan , D. Frank , R. Kurt , M. Christian , Towards virtualizationconcepts for novel automotive hmi systems, in: Embedded Systems: Design,

Analysis and Verification, in: IFIP Advances in Information and Communica-

tion Technology, 2013, pp. 193–204 . [133] Z. Gu, Z. Wang, S. Li, H. Cai, Design and implementation of an automotive

telematics gateway based on virtualization, in: Object/Component/Service-Oriented Real-Time Distributed Computing Workshops (ISORCW), 2012 15th

IEEE International Symposium on, 2012, pp. 53–58, doi: 10.1109/ISORCW.2012.20 .

[134] R. Ponmagal , N. Dinesh , U. Rajaram , Design and development of secure cloud

architecture for sensor services, in: Distributed Computing and Internet Tech-nology, in: Lecture Notes in Computer Science, 2015, pp. 339–344 .

[135] P. Nayot , K. Vimal , M. Sanjay , C. Sriram , Challenges in secure sensor-cloudcomputing, in: Secure Data Management, in: Lecture Notes in Computer Sci-

ence, 2011, pp. 70–84 .

R. Chaâri et al. / Computer Networks 108 (2016) 260–278 277

er science in the National School of Computer Science and Engineering (ENSI). She is a

Systems) research group. She received her master degree from the National Engineer- the performance evaluation of 6LoWPAN networks behavior and RPL protocol as well.

for robotic applications. Research Interests : Cloud computing, cloud robotics, cooperative

uter Engineering and Applied Mathematics (DGIMA) in National School of Engineers of

Intelligent Networked Systems) research group. She received her engineer diploma from

is project focuses on virtualizing robotic services using cloud computing concepts.

ngineering from Higher School of Telecommunications (Tunisia), and M.Sc. degrees in

), in 20 0 0 and 20 01, respectively, and the Ph.D. degree in Computer Science from the 4. He was a faculty member at Al-Imam University from 2006 to 2012. Currently, he is

ience at Prince Sultan University and research associate in CISTER Research Unit, ISEP- Education Academy (SFHEA) in 2015. He has published over 120 refereed journal and

obots, robotics software engineering, Internet-of-Things, cloud computing and wireless rd from Al-Imam University in 2010, and the best paper award of the 19th Euromicro

he head of the ACM Chapter in Prince Sultan University. His H-Index is 27.

llege of Computer & Information Science at Prince Sultan University. He received his PhD UK and Master of Science degree in Computer Science from Florida Atlantic University

rks Security, Trust security and privacy (TSP) in wireless networks and TSP in P2P appli-

as published in various reputable international journals and conferences. He is a member ty and ACM.

of the Polytechnic of Porto (ISEP). The research by Nuno Pereira was published in more es. He was involved in numerous international research projects and was the Principal

ireless sEnsoR Networks), and sub-Project Leader and the High-level Architect in DEWI Nuno served as the Technical Program Co-Chair of EWSN 2015.

rmatique from the Faculté des Sciences de Tunis, University of Tunis El Manar, Tunisia in sity of Minnesota Twin Cities, USA, in January 1990. He is a Professor of computer science

sse. Since October 2013, he has been serving as the Director General of Centre de Calcul

nisian National Research and Education Network and providing Internet and application f has over 200 publications in the form of books, book chapters, journal and conference

rks, performance evaluation of computer systems, and combinatorial optimization.

Rihab Chaâri is a Ph.D student in department of comput

member of The COINS (Cooperative Intelligent Networkeding School of Sfax in 2011. Her master thesis versed on

His current work focuses on designing cloud architecture

robots.

Fatma Ellouze is a PhD student in Department of Comp

Sfax (ENIS). She is a member of The COINS (Cooperative the National School of Engineers of Sfax in 2013. Her thes

Anis Koubaa received his B.Sc. in Telecommunications E

Computer Science from University Henri Poincare (FranceNational Polytechnic Institute of Lorraine (France), in 200

an associate professor in the Department of Computer ScIPP, Portugal. He becomes a Senior Fellow of the Higher

conference papers. His research interest covers mobile rsensor networks. Dr. Anis received the best research awa

Conference in Real-Time Systems (ECRTS) in 2007. He is t

Basit Qureshi is currently an assistant professor in the Codegree in Computer Science from University of Bradford,

respectively. His research interests include Wireless Netwo

cations for Mobile Ad Hoc wireless Networks. To date he hof IEEE, IEEE Computer Society, IEEE Communication Socie

Nuno Pereira is a Professor at the School of Engineering than 30 technical papers in peer reviewed scientific venu

Investigator of PATTERN (Programming AbsTracTions for w(DEpendable Embedded Wireless Infrastructure). Recently,

Dr. Habib Youssef received a Diplôme d’Ingénieur en InfoJune 1982 and a Ph.D. in computer science from the Univer

at the ISITCom Hammam Sousse of the University of Sou

El-Khawarizmi, Tunis, Tunisia, which is managing the Tuservices to the Tunisian academic community. Dr. Yousse

papers. His current research interests are computer netwo

278 R. Chaâri et al. / Computer Networks 108 (2016) 260–278

l and computer engineering from the University of Porto, Porto, Portugal. Since 2003 he

tute of Porto (ISEP-IPP), an internationally renowned research centre focusing on RTD in ly engaged in research on real-time distributed systems, multiprocessor systems, cyber-

e is currently the Vice-chair of ACM SIGBED (ACM Special Interest Group on Embedded

2015, member of the Executive Committee of the IEEE Technical Committee on Real-Time an 150 scientific and technical papers in the area of real-time and embedded computing

tributed embedded systems. He has been consistently participating in top-rated scientific as been program chair/co-chair for ECRTS 2005, IEEE RTCSA 2010, IEEE RTAS 2013 or IEEE

s. He has also been program chair/co-chair of other key scientific events in the area of stems as is the case of ARCS 2014 or the ACM/IEEE ICCPS 2016.

Eduardo Tovar holds since 1999 a PhD degree in electrica

heads the CISTER Research Centre of the Polytechnic Instireal-time and embedded computing systems. He is deep

physical systems and industrial communication systems. H

Computing Systems) and was for 5 years, until December Systems. Since 1991 he authored or co-authored more th

systems, with emphasis on multiprocessor systems and disevents as Program Chair or as General Chair. Notably he h

RTCSA 2016, all in the area of real-time computing systemarchitectures for computing systems and cyber-physical sy