ArticlePDF Available

Live Data Analytics With Collaborative Edge and Cloud Processing in Wireless IoT Networks

March 2017
IEEE Access PP(99):1-1

March 2017
PP(99):1-1

DOI:10.1109/ACCESS.2017.2682640

License
CC BY-NC-ND 4.0

Authors:

Shree Krishna Sharma

Xianbin Wang

The University of Western Ontario

Recently, big data analytics has received important attention in a variety of application domains including business, finance, space science, healthcare, telecommunication and Internet of Things (IoT). Among these areas, IoT is considered as an important platform in bringing people, processes, data and things/objects together in order to enhance the quality of our everyday lives. However, the key challenges are how to effectively extract useful features from the massive amount of heterogeneous data generated by resource-constrained IoT devices in order to provide real-time information and feedback to the end-users, and how to utilize this data-aware intelligence in enhancing the performance of wireless IoT networks. Although there are parallel advances in cloud computing and edge computing for addressing some issues in data analytics, they have their own benefits and limitations. The convergence of these two computing paradigms, i.e., massive virtually shared pool of computing and storage resources from the cloud and real-time data processing by edge computing, could effectively enable live data analytics in wireless IoT networks. In this regard, we propose a novel framework for coordinated processing between edge and cloud computing/processing by integrating advantages from both the platforms. The proposed framework can exploit the network-wide knowledge and historical information available at the cloud center to guide edge computing units towards satisfying various performance requirements of heterogeneous wireless IoT networks. Starting with the main features, key enablers and the challenges of big data analytics, we provide various synergies and distinctions between cloud and edge processing. More importantly, we identify and describe the potential key enablers for the proposed edge-cloud collaborative framework, the associated key challenges and some interesting future research directions.

Main attributes of big data

…

Main technology enablers for big data analytics

…

Proposed generalized system model for collaborative edge-cloud processing in heterogeneous IoT networks

…

Figures - uploaded by Shree Krishna Sharma

Content may be subject to copyright.

Content uploaded by Shree Krishna Sharma

Content may be subject to copyright.

Available via license: CC BY-NC-ND 4.0

Content may be subject to copyright.

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2682640, IEEE Access

IEEE ACCESS (ACCEPTED FOR PUBLICATION) 1

Live Data Analytics with Collaborative Edge and

Cloud Processing in Wireless IoT Networks

Shree Krishna Sharma, Member, IEEE, Xianbin Wang, Fellow, IEEE

Abstract—Recently, big data analytics has received important

attention in a variety of application domains including business,

ﬁnance, space science, healthcare, telecommunication and Inter-

net of Things (IoT). Among these areas, IoT is considered as

an important platform in bringing people, processes, data and

things/objects together in order to enhance the quality of our

everyday lives. However, the key challenges are how to effectively

extract useful features from the massive amount of heterogeneous

data generated by resource-constrained IoT devices in order to

provide real-time information and feedback to the end-users,

and how to utilize this data-aware intelligence in enhancing

the performance of wireless IoT networks. Although there are

parallel advances in cloud computing and edge computing for

addressing some issues in data analytics, they have their own

beneﬁts and limitations. The convergence of these two computing

paradigms, i.e., massive virtually shared pool of computing and

storage resources from the cloud and real-time data processing

by edge computing, could effectively enable live data analytics

in wireless IoT networks. In this regard, we propose a novel

framework for coordinated processing between edge and cloud

computing/processing by integrating advantages from both the

platforms. The proposed framework can exploit the network-

wide knowledge and historical information available at the

cloud center to guide edge computing units towards satisfying

various performance requirements of heterogeneous wireless

IoT networks. Starting with the main features, key enablers

and the challenges of big data analytics, we provide various

synergies and distinctions between cloud and edge processing.

More importantly, we identify and describe the potential key

enablers for the proposed edge-cloud collaborative framework,

the associated key challenges and some interesting future research

directions.

Index Terms— Big data, Data analytics, Internet of Things

(IoT), Cloud computing, Edge computing, Fog computing.

I. INTRODUCTION

The current trend in the Internet world is to connect all

the devices/objects/things to the Internet with the objective of

enhancing the quality of our everyday lives, thus leading to

the emergence of Internet of Things (IoT) [1], [2]. In this

direction, there has been a tremendous growth in the number

of Internet-enabled smart devices and connections such as

smartphones, Machine to Machine (M2M) connections, smart

home appliances, and smart wearable devices, and this trend

is expected to continue in the future. According to CISCO,

more than 50 billion devices are expected to be connected to

the Internet by 2020. Recent advances in sensing, computing,

wireless communications, Internet protocols, and networking

technologies have made the concept of IoT feasible [1].

S. K. Sharma and X. Wang are with Department of Electrical and Computer

Engineering, Western University, 1151 Richmond St, London, Ontario, N6A

3K7, Canada, Email: {sshar323, xianbin.wang}@uwo.ca.

However, the main challenge is how to handle the real-time

processing of a huge amount of data/information, called big

data, generated from heterogeneous wireless IoT environment.

Big data may be generated from various environments such

as e-Healthcare environment, online business/e-commerce,

broadband and multimedia contents, cloud radio access net-

works, and distributed storage/sensing [3]. Besides the content

and trafﬁc-related data, there is continuously increasing vol-

ume of signaling data due to the rapid deployment of various

wireless networks including mobile and IoT networks. The

complexity of big data generated from the IoT environment

depends on the computational cost required in processing the

data rather than the size of data itself. Besides, this massive

amount of data needs to be transferred from the edge nodes

to the cloud, leading to the need of enormous communication

bandwidth which is precious and expensive natural resource.

In addition, this massive data needs to be stored for further

processing and also to facilitate real-time delivery at the edge-

side, thus leading to the storage/caching constraints.

Existing wireless networks are mainly designed by consid-

ering communication resources as the primary resources with

the connection-oriented approach, and other resources such

as computing and caching are considered as secondary [4].

However, the demand is towards content-oriented networks

facilitated by the recent advances in Internet technologies and

cloud computing, where computing and caching will be the

integral parts of the network. Also, in the ﬁfth generation

(5G) and beyond wireless networks, all types of resources such

as communication, computing and caching will be distributed

throughout the network, and it is an important challenge

to coordinate among these resources towards their effective

utilization in handling the massive amount of distributed data.

One of the promising approaches to tackle the issues of the

big data could be to enable synergies among communications,

computing and caching components of future wireless IoT

networks [3]. The integration of these paradigms may lead

to additional degrees of freedom in effectively optimizing

the resources of communication systems. In this regard, it

has become an essential requirement to take all the involved

resources into account while designing future wireless IoT

networks by exploiting the synergy among communications,

caching and computing paradigms [4], [5].

One of the recent developments in the computing world is

the Internet-based computing, called cloud computing, which

provides an ubiquitous and on-demand access to a virtually

shared pool of conﬁgurable computing and storage resources

[6]. Cloud computing is an excellent platform to handle the

enormous data generated from the IoT environment due to

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

cheaper and large amount of virtual computing/processing

power available at the cloud centre. Therefore, the current

trend is towards IoT-cloud convergence with most of the IoT

platforms supported with cloud computing. However, it is not

suitable for the applications demanding low-latency, real-time

operation and high Quality of Service (QoS) [7]. In addition to

low latency and location awareness requirements, the emerging

IoT platform requires the support for seamless mobility and

ubiquitous coverage which can not be fully supported by cloud

computing solutions.

On the other hand, the concept of edge computing, also

called fog computing1, is receiving important attention in

order to address some of the drawbacks of cloud computing

[8]–[10]. The main goal of edge computing is to extend the

cloud computing functions to the edges of the network. Due

to proximity to the end-users and geographically distributed

deployment, it can support the applications/services demand-

ing the requirements of low-latency, location-awareness, high

mobility and high QoS [9]. However, edge computing units

usually do not have enough storage and computing resources

in handing the massive amount of IoT data. In addition, due to

several involved constraints such as low-power, heterogeneity

and weak capability of devices, IoT environment is more

vulnerable to the information security. Therefore, there is a

clear need to investigate suitable network architecture and

control mechanisms to handle the processing of massive IoT

data in a secured manner.

Although there are ongoing parallel advances in the ﬁelds

of cloud computing and edge computing, interactions between

these platforms in handling live data analytics from the

communication perspective have not been investigated in the

literature. A few works have recently highlighted the need of

coordination between edge computing and cloud computing

[8], [11], [12], however, they do not consider various practical

aspects of live data analytics in wireless IoT networks. In this

regard, this paper proposes a novel framework of collaborative

edge-cloud processing in order to handle live data analytics in

wireless IoT networks by combining advantages from both the

cloud and edge computing paradigms. Due to huge amount

of storage and computing resources available at the cloud-

end, it is beneﬁcial to ofﬂoad much of the computational

tasks to the cloud. However, it becomes highly advantageous

to handle delay-sensitive tasks at the edge-side in order to

ensure real-time processing and feedback to the end-users.

More importantly, in the proposed framework, cloud pro-

cessing can utilize its network-wide global knowledge and

historical/delayed information and can act as a monitoring

or guidance platform to guide edge processing units towards

the effective-utilization of available resources. Starting with

the basic features and key enablers of big data analytics, we

discuss various challenges in performing live data analytics

in wireless IoT networks. Subsequently, we provide synergies

and differences between cloud and edge computing platforms.

Then, we propose a novel framework for collaborative pro-

cessing between edge and cloud computing along with some

1In this paper, the terms “edge computing” and “fog computing” are

used interchangeably, and they refer to the computing/processing at the IoT

gateway/aggregator nodes/edge servers.

interesting applications. Subsequently, we discuss various key

enablers, associated challenges and future research directions

with the objective of stimulating future research activities in

this emerging domain.

The remainder of this paper is organized as follows:

Section II provides the basic features, the key technology

enablers for big data analytics and the associated challenges in

wireless IoT networks. Section III compares edge processing

and cloud processing platforms from the perspective of live

data analytics. Section IV proposes a novel collaborative edge-

cloud processing framework for handling live data analytics

in wireless IoT networks while Section V discusses various

technology enablers, issues and future research directions.

Finally, Section VI concludes the paper.

II. DATA ANALYTICS IN WIRELESS IOT NETWORKS

In this section, we provide the basic features, key enablers

and the challenges for big data analytics in wireless IoT

networks.

A. Basic Features

The term “Big data” usually refers to extremely large,

heterogeneous and complex (semi-structured and unstructured)

data-sets, which cannot be handled by the conventional data

processing and storage tools/applications such as Relational

Database Management System (RDBMS) [13]. The impor-

tance of big data lies on how meaningful information can

be extracted from it for a particular application rather than

the size of the data, and this extraction process requires novel

data analysis methods and huge processing power. In wireless

IoT environments, big data may be generated from a variety

of application scenarios ranging from smart home scenario

to e-Healthcare applications. In addition to the importance

of content and control signaling data in wireless networks,

location-based data from various sensors such as GPS sen-

sors and embedded sensors in mobile devices can provide

signiﬁcant inputs to the government bodies in developing

speciﬁc strategies for public facilities, transportation system,

emergency responses and crime/risk warnings. Moreover, by

analyzing the habits and interests of customers, industries may

plan their future products in order to address their customers’

personalized as well as group needs [3].

As depicted in Fig. 1, the commonly discussed attributes

of big data are [13]: (i) volume, (ii) variety, (iii) veracity, (iv)

velocity, and (v) value. The ﬁrst two attributes, i.e., volume

and variety, reﬂect to the hardware and software requirements

in handling massive heterogeneous data-sets while the variety

and velocity translate into the real-time processing ability with

sufﬁcient trustworthiness. On the other hand, acquisition of

the highest useful value from the complex big data-sets in

wireless IoT networks requires interdisciplinary cooperation

among academia, enterprises and wireless industries [13].

B. Challenges

In contrast to the traditional data, big data mainly differs

in the following way [14]: (i) data rate is more rapid and data

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Big Data

Features

Volume

Velocity

Veracity

Variety

Value

Fig. 1. Main attributes of big data

volume is constantly updated, (ii) data is of semi-structured

or unstructured nature, (iii) data source is fully distributed,

(iv) data access is in batch mode or real time instead of more

interactive feature in the traditional data, and (v) integration

of heterogeneous data from different sources becomes compli-

cated. The heterogeneous data generated from IoT devices may

have certain statistical and strong correlative features across

several dimensions such as time and location, and also the

devices may have social relations among themselves [15]. In

addition, in hierarchial IoT networks, the aggregated features

of the data trafﬁc can be exploited in order to regulate the

peak content demand, for example, cluster planning based on

data distribution, peak load shifting and cache provisioning.

Besides, IoT devices with similar interests may share the con-

tents from their nearby devices and this content sharing may

be enabled using either infrastructure-based communication or

some infrastructure-free communication. Such a peer to peer

nature of resource sharing, called as crowd computing [15],

can exploit the spatial correlation as well as the mobility of

IoT devices.

Big data analytics at the cloud centre can easily integrate

the data collected using distributed sensors and aggregator

nodes while exploiting correlation among the data-sets [16].

More importantly, this has the ability to analyze the mas-

sive data with ever-increasing scale and complexity, and can

provide a global point of view across the whole network.

Moreover, this cloud-based approach will lead to lower-error,

higher-precision, and more dynamic treatment of data than the

conventional data analytic approaches [16]. However, handling

IoT data in the cloud platform in a traditional way creates

several issues due to speciﬁc features of IoT data described in

the following [17].

1) Distributed and heterogeneous data structure: IoT

data is generated from distributed heterogeneous nodes

which is largely diverse and may range from integer to

character, and can be semi-structured and unstructured

such as audio, images, and video. In addition, big data

in wireless networks is usually distributed across several

domains such as frequency, space, time, codes, and

antennas. Also, the involved data sources may have

distinct characteristics in terms of data rate, mobility,

power levels and transmission schemes.

2) Real-time requirements: The dynamic environment of

wireless IoT systems creates the need to handle a large

volume of real-time and high-speed continuous data

streams.

3) Weak data semantics: The data acquired from IoT

sensors are mainly of low-level having weak semantics

and are gathered with the help of resource-constrained

sensors/devices/objects. In order to extract meaningful

information from the collected data, they need to go

through effective processing by exploiting various as-

pects such as spatial-temporal correlation and event-

driven knowledge.

4) Data inaccuracy: Since the information gathered by

the employed sensing system may not be accurate

enough due to various practical constraints, suitable

multi-dimensional sensing, data analysis and processing

techniques need to be investigated for wireless IoT

applications.

The data analytics life-cycle to handle IoT data starts with

the collection of the raw data collection at the IoT sensors,

and then the acquired raw information is aggregated at the

aggregator/gateway, and is subsequently sent to the cloud for

further processing [17]. The raw information gathered from

the IoT sensors may represent various parameters such as

vibration, pressure, temperature, motion, heart rate etc., and

they need to be converted into recognizable formats for further

analysis. The acquired data may be transmitted to the cloud

centre either in a single hop or multiple hops based on the

employed network topology. The completion of this life-cycle

requires effective coordination of various activities among

IoT sensors, IoT aggregator nodes/gateways, in-transit network

devices such as routers/relays and software/hardware resources

in the cloud center [18]. Furthermore, it is crucial to have

reliable and effective wireless communication links among

these entities in order to ensure proper operation of wireless

IoT networks.

Another key issue for handling big IoT data is how to turn

the burden of handling big data to the beneﬁts in improving the

performance of wireless networks [15]. The big data usually

exhibits highly useful features such as user activity/mobility

patterns and temporal, spatial and social correlations of data

contents. By properly extracting and effectively utilizing these

features, the performance of various wireless networks could

be signiﬁcantly enhanced. For example, by analyzing social

ties and common interests of people in a certain region, the

network can fetch the popular contents to the corresponding

edge gateway in order to reduce latency as well as the

communication bandwidth overhead.

In wireless IoT networks, it is crucial to develop tech-

niques to convert a massive amount of raw data into the

meaningful information. The main challenge during this data

interpretation and knowledge formation is to develop suitable

data inference techniques to deal with the noisy and real-

world data. Furthermore, dealing with the uncertainty in the

interpreted data is another challenge due to dynamics of time-

varying wireless environment. In order to make a reliable and

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Enablers for big data

analytics

Stochastic

modeling

Distributed

and parallel

optimization

Supervised

learning

Random

matrix theory

Machine

learning

Advanced

learning

techniques

Data Mining

Cloud

computing

Edge

computing

Processing

methods

Batch

processing

techniques

Stream

Processing

techniques

Caching and

offloading

techniques

Computing

platforms

Active

learning

Deep

learning

Online

learning

Reinforceme

nt learning

Markov

models Time series,

geometric

models

Kalman

filters

Unsuper vised

learning

ing

Crowd

computing

SDN, Dat a

virtualization

Fig. 2. Main technology enablers for big data analytics

trustworthy decision, reliable transport protocols and suitable

in-ﬁeld sensor calibration techniques need to be investigated

[19]. Otherwise, it may lead to wrong or incomplete data,

resulting in false conclusions, which if used in practical

scenarios may lead to serious problems. In order to tackle this,

the inferred information can be supported by a probability-

based conﬁdence level which can be used to ensure the safe

operation of the devices [19]. Moreover, in several scenarios,

it may be necessary to combine the current sensor data with

the historical data in order to derive an effective conclusion.

C. Enablers for Big Data Analytics

The conventional data management tools such as RDBMS

are mainly designed to handle structured data and are not

suitable for handling semi-structured or unstructured data. In

addition, for handling the massive IoT data, RDBMSs can

scale up with the costly hardware but not with the commodity

hardware [14]. To address the aforementioned issues, some

ad-hoc solutions have been recently proposed by the research

community. In terms of infrastructure level, cloud computing

has been considered important for handling big data due to

its several features such as scalability, elasticity and cost-

effectiveness. In order to handle massive unstructured data-

sets, NoSQL databases and distributed ﬁle systems have

been suggested. Furthermore, programming frameworks like

MapReduce have been proposed in order to handle group-

aggregation tasks like website ranking. Moreover, in terms of

system-level solution, open-source software frameworks like

Hadoop have been proposed to integrate data storage, data

processing and other modules [14]. However, these solutions

are not mature enough in order to handle huge data collection

and transmission, and to provide real-time processing and

feedback to the end-users. Also, high power consumption,

large network overhead, latency, privacy and security are of

important issues to be addressed.

Figure 2 illustrates the key technology enablers for big data

analytics. In the following, we provide brief description of

these techniques from the data analytics perspective. Stochas-

tic models are probabilistic models and are usually used to

capture the explicit features and dynamics of the data trafﬁc.

The commonly used stochastic models are Markov models,

time series, geometric models, and Kalman ﬁlters [15]. On

the other hand, data mining approach tries to extract implicit

information from the data-sets and transform this information

to a known structure for further usage by employing suitable

anomaly detection, classiﬁcation, clustering, and regression

analysis methods.

Machine learning techniques aim to create a functional

relationship between input data-set and output actions, and

are capable of performing predictions and decisions based

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

on the input data without requiring the need of following

static program instructions. These techniques can be broadly

categorized into unsupervised, supervised and reinforcement

learning, and they may comprise of various classiﬁcation

techniques, regression analysis and Q-learning techniques. In

addition to the conventional machine learning techniques,

several advanced learning techniques such as active learning,

deep learning, and online learning can be utilized to extract

useful information from incomplete or complex data-sets.

Active learning is mostly useful for partially labeled data-

sets while deep learning is suitable for modeling complex

behaviors of heterogeneous data-sets [15]. On the other hand,

online learning deals with learning in real-time and is useful

for applications where data arrives in a sequential order.

In terms of computing platforms, edge computing and

cloud computing are considered as key solutions for handling

big data analytics, and they will be detailed later in Section III.

In addition, crowd computing is another emerging paradigm

in which nearby devices with social ties and similar interests

can share resources in order to maximize the overall system

performance. By implementing this paradigm, devices having

higher computing capability and power resource can help

other devices with lower computing capability and battery

level in achieving their performance targets. Moreover, crowd

computing-based solutions can exploit temporal, spatial cor-

relations as well as mobility of IoT devices for informa-

tion/resource sharing and can subsequently reduce the back-

haul/fronthaul bandwidth in infrastructure-based wireless IoT

networks [15].

Based on the processing time required for big data an-

alytics, the potential techniques can be broadly classiﬁed

into stream processing and batch processing [14]. In the ﬁrst

approach, the data arrives to the system continuously in the

form of a stream and the data size is inﬁnite or unknown in

advance. Whereas in the batch processing method, data size is

ﬁnite and known in advance, and they are stored and analyzed.

In this approach, the data are divided into small chunks and

processed in parallel in a distributed manner.

The widely used linear algebra methods such as Cholesky

decomposition and matrix multiplications, and convex opti-

mization algorithms can not be straightforwardly applied for

big data analytics because of extremely large sizes of data

as well as the involved parameters [20]. This requires the

need of adapting the existing algorithms and investigating

new optimization techniques for big data analytics problems.

Optimization algorithms suitable for big data can be catego-

rized into the following types [20]: (i) ﬁrst order methods, (ii)

randomization, and (iii) parallel and distributed computation.

Out of these, distributed optimization algorithms play a key

role in solving resource allocation and management problems

in heterogeneous wireless IoT networks where information

and resources are often distributed across various entities

of the networks. Some examples of distributed optimization

algorithms are alternating direction method of multipliers and

primal/dual decomposition, which can decompose complex

optimization problems into sub-problems in such a way that

they can be simultaneously computed [15]. This parallel

computation capability signiﬁcantly helps in reducing com-

putational burden in a single computing unit as well as the

communication overhead over the fronthaul/backhaul.

One possible way of dealing with the high dimensionality

of big data is to use random matrix theory [21], [22] by rep-

resenting big data in the form of large random matrices. This

analysis is based on some of large dimensional matrix analysis

tools such as high-dimensional statistics, matrix analysis, and

convex optimization [16]. For example, a massive Multiple

Input Multiple Output (MIMO) system can be regarded as a

big data system considering storage and processing at the IoT

gateway, and the principles of large random matrices can be

applied to the architecture of large antenna array of massive

MIMO system [23]. In addition, several dimension reduction

approaches such as Principal Component Analysis (PCA) and

tensor decomposition can be employed in order to reduce the

data volume without changing the main features of the data

[15]. By reducing the dimension of data, a signiﬁcant gain can

be achieved in saving the system cost for storage, processing

and communication resources. As illustrated in Fig. 2, other

enablers of big data are caching techniques [24], ofﬂoading

techniques [25], Software Deﬁned Networking (SDN) [26] and

virtualization technologies [27], which will be described later

in Section V.

III. EDGE COMPUTING VERSUS CLOUD COMPUTING

Existing wireless networks and Cloud/Centralized Radio

Access Networks (C-RAN) under investigation are mainly

designed to deliver contents without having the capability

of analyzing and making use of the data-speciﬁc features

in optimizing the system performance [15]. In this regard,

it is important to make existing wireless networks scalable

in order to handle massive data contents. Also, it is crucial

to investigate suitable network architectures/mechanisms to

incorporate and utilize big data awareness in wireless networks

in order to enhance the system performance. As the amount

of data generated by a large number of distributed sensors is

highly unstructured and heterogeneous in nature, it becomes

extremely complex to handle them with the conventional

approaches. Mainly, how to acquire, integrate, store, process

and utilize the data in highly distributed environments has

been an important challenge for researchers, engineers and

data scientists.

To address these issues, CISCO has recently proposed

the concept of fog computing which aims to support cloud

computing platform in handling a part of the workload locally

at the edge devices such as switches, routers, and IP-enabled

video cameras instead of transmitting the whole work load to

the cloud [28]. The fog/edge computing can be enabled in the

existing cloud-based networks by introducing an intermediate

layer, called fog/edge layer, which may comprise of several

edge servers distributed over various places such as shopping

centres, parking areas and bus stations. The edge server can

be regarded as a low-capacity version of the cloud server and

has communication, computing and data storage capabilities.

Edge computing becomes highly advantageous for mobility

support, geo-distribution, and location/context awareness. The

geo-distributed nature of edge computing helps to provide

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

TABLE I

KEY DIFFERENCES BETWEEN EDGE COMPUTING AND CLOUD COMPUTING

Features

Cloud computing

Edge computing

Computational capacity

High

Medium to low

Size and Operating mode

Server very large in size and

centralized

Edge servers smaller in size and

placed over many locations

Applications

Suitable for delay-tolerant and

computationally

-intensive

applications

Suitable for applications demanding

low latency, real

time operation and

high QoS

Fronthaul/backhaul

communication overhead

High since devices need to be

connected to Internet throughout

entire duration

Low since devices can get cached

contents directly from edge

gateway

Deployment

Requires complicated deployment

planning

Possibility of ad-hoc deployment

with no or minimal planning

rich contextual information such as event status, local net-

work conditions, and the end-user’s status. These information

can be subsequently used for context-aware optimization of

edge/fog applications. More speciﬁcally, edge computing can

support the existing cloud computing platform in handling the

following different types of applications, whose requirements

can not be met with the cloud processing [28].

1) Applications demanding very low and predictable la-

tency such as video conferencing, online gaming, and

e-Healthcare

2) Real-time mobile applications such as smart connected

vehicles

3) Geographically distributed applications such as wireless

sensor networks for environmental monitoring

4) Large-scale distributed control systems such as smart

trafﬁc lights, smart grid and smart energy distribution

As mentioned earlier, the main beneﬁts of cloud computing

platform in wireless IoT networks are massive storage, very

high computational efﬁciency, wide-area coverage while the

main advantages of edge computing are real-time data han-

dling, edge resource pooling, user-centric process, the support

for high mobility and high QoS [7], [12]. In Table I, we

provide the key differences between edge computing and cloud

computing platforms from the perspective of handling big-data

in wireless IoT networks [12].

Due to ever-increasing demand for data content, the trans-

mission of massive amount of data to the cloud creates a

huge burden on the communication bandwidth of wireless

networks. Furthermore, this results in intolerable latency and

degraded service to the end-users. Moreover, the support

of geo-distribution and mobility is an essential requirement,

which can not be fulﬁlled by the cloud computing platform due

to its centralized nature of storage and computing functions

[28]. Therefore, it is crucial to explore the collaborative

processing of edge and cloud computing platforms, which will

be described in the following section.

IV. PROPOSED COLLABORATIVE EDGE-CLOUD

PROCESSING

As highlighted in Section III, edge computing and cloud

computing solutions have their own distinct advantages and

disadvantages from the perspective of live data analytics in

wireless IoT networks. The integration of centralized feature of

the cloud and the real-time advantage of edge computing can

address various issues in dealing with real-time data analytics

in wireless IoT networks. Motivated by this aspect, in this

section, we propose a novel framework for collaborative edge-

cloud processing in wireless IoT networks.

Figure 3 presents a generalized system model for col-

laborative edge-cloud processing in heterogeneous wireless

IoT networks. In the proposed model, IoT edge gateways are

equipped with cache memory and are capable of performing

edge-caching in order to deliver the popular contents locally.

The edge computing nodes may be any devices having the

capability of computing, storage and network connectivity

such as routers, switches, and video surveillance cameras.

Depending on the application scenarios, IoT networks may

comprise of various networks having distinct characteristics.

For example, in the smart home scenario, wireless IoT net-

works may consist of a WiFi network, a blue-tooth network, a

Zigbee network and a cellular network. The raw-data coming

from different domains/sensors is largely diverse and need to

be collected over time. In addition, data dimensions and sizes

may be different depending on the considered IoT application

scenario. Besides the real-time processing of massive IoT

data, this collaborative framework an enable new wireless IoT

applications which may require collaborations among different

edge computing units, and between edge computing units and

the cloud centre.

The proposed system will beneﬁt from the advantages of

both the cloud computing and edge computing. In addition to

this, we envision cloud centre as a monitoring and guidance

platform to have effective real-time data processing at the

edge-side of wireless IoT networks. In practical scenarios,

IoT devices/sensors are heterogeneous in nature in terms

of their computing capabilities, intelligence as well as the

computing/processing power. In this regard, it becomes highly

beneﬁcial to guide the operation/processing of edge-nodes in

order to utilize the available communication and computing

resources in an effective manner. In the considered framework,

edge computing helps to gather information from the surround-

ing radio environment while the cloud computing assists by

providing suitable instructions to the edge-side nodes for their

operations. For example, the operations at the edge-side such

as data compression, ﬁltering, sampling rate, power control,

and making decisions on the type of data to be sensed/acquired

can be supported by the cloud centre by providing suitable

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

WiFi

ZigBee Cellular

IoT

Gateway

Core network/

Internet cloud

Backhaul

link

Fronthaul

links Cache

Edge processing

Cloud processing

Interactions

Fig. 3. Proposed generalized system model for collaborative edge-cloud processing in heterogeneous IoT networks

control signals over the feedback links.

Since the cloud centre can have a global view of informa-

tion collected from a large number of sensors deployed over a

large geographical region, the control of edge processing from

the cloud-side can provide signiﬁcant improvements in future

wireless IoT networks. Due to huge amount of computing

resources available at the cloud end, it is beneﬁcial to ofﬂoad

much of the computational tasks to the cloud. On the other

hand, it is advantageous to handle delay-sensitive tasks at the

edge-side. Depending on various levels of information such

as trafﬁc types, location information, processing delay and

transmission overhead, the decision on whether to ofﬂoad data

to the cloud or not can be made. It can also be considered that

all the edge nodes are operated in a coordinated fashion in

order to help each other in terms of communication, comput-

ing and storage/caching resources. Another important aspect

which can be exploited in the proposed framework is that

cloud processing can utilize the history/delayed information

available at the cloud-centre in order to infer certain decisions

for the edge processing without the need of waiting for the

instantaneous data collected from IoT nodes.

The introduction of edge-computing in the conventional

centralized cloud computing set-up brings up new opportuni-

ties to balance the trade-off between centralized and distributed

network architectures [12]. In this regard, the proposed col-

laborative edge-cloud processing will be signiﬁcantly useful

to decide on what actions to perform locally and what actions

to be sent to the cloud. Various constraints in the considered

system model include computational rate, computing power,

processor speed, cache size, communication bandwidth, la-

tency and transmit power. The performance of the considered

framework in Fig. 3 can be evaluated in terms of various met-

rics such as energy efﬁciency, spectral efﬁciency, throughput,

operational efﬁciency, cache hit ratio, computational efﬁciency,

end to end latency and ofﬂoading efﬁciency.

Since wireless environment is time varying in nature, it is

crucial to adapt the proposed collaborative processing dynam-

ically. Depending on different network conditions, physical

channel conditions and data features such as temporal and

spatial correlations, the proposed collaborative platform may

adaptively choose among central processing at the cloud

centre, local processing at the edge gateway and parallel

processing at both the ends. The historical data available at

the cloud side and the instantaneous data collected by the edge

nodes at the current instance can be used to predict various

important parameters such as future trafﬁc ﬂow, energy usage,

weather forecasting, community activity, and geo-location. In

this context, it is important to investigate dynamic prediction

algorithms which can be updated based on time-varying situ-

ations such as source node failure and security threat.

Due to massive amount of IoT data and continuously

increasing number of service requests, power consumption

for operating servers in the cloud centre is rapidly increasing

[29]. Therefore, it is crucial to investigate suitable strategies

to reduce energy consumption in the edge-cloud coordinated

platform. On the other hand, it is important to guarantee the

latency requirements while delivering services to the end-

users. In this regard, it is essential to investigate the funda-

mental tradeoff between energy consumption and latency in

the considered cloud-edge platform of wireless IoT networks

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

[28]. Moreover, the massive data-sets gathered from various

locations have heterogeneous formats and are usually semi-

structured and unstructured in nature. In most cases, the data-

sets are in a raw form with inconsistency, high redundancy and

much useless information. Without performing pre-processing

operations on the data-sets, not only a huge storage is required

but also the data may not ﬁt into the predeﬁned database

structure. Therefore, it becomes highly beneﬁcial to carry

out data pre-processing such as data integration, redundancy

elimination and data cleansing at the edge-side in order to

avoid the unnecessary storage space, and also to enhance the

computational efﬁciency.

Furthermore, it may not be necessary for all IoT sensor

nodes to upload the collected/generated data to the cloud. In

this regard, the edge computing unit can inform IoT devices

when to stop sending the data to the cloud in order to

effectively utilize the network resources. In addition, a certain

portion of the data may not be required for a certain time

period. In this case, the IoT gateway can inform the IoT

sensor/device when it needs to stop uploading the data. This

collaborative procedure will help to save the cloud as well

as the network resources for that idle period. Besides, in the

existing scenarios, many nodes are directly connected to a sink

node/access point/coordinator, which leads to signiﬁcant in-

crease in the complexity for node scheduling, and subsequently

it will lead to the intolerable system delay. In this context, the

proposed framework follows a hierarchical network structure

in which the nodes can be grouped locally into different

clusters and they can be connected to the sink via their cluster

heads [30]. By enabling collaborations among the devices

within a cluster using their social relations, devices can share

various communication, storage and computing resources, and

subsequently the network performance can be enhanced in

terms of different performance metrics such as latency and

energy efﬁciency, content delivery efﬁciency and security.

The proposed collaborative edge-cloud processing frame-

work can be applied to handle real-time data analytics in

wireless IoT networks with different performance objectives.

Some of the potential applications are listed below.

•Cloud-assisted adaptive optimization of computing, com-

munications and caching resources

•Cloud-assisted energy-efﬁcient caching and task/data of-

ﬂoading

•Spectrum monitoring and dynamic spectrum management

using collaborative edge-cloud processing

•Event-driven resource allocation and network manage-

ment using collaborative edge-cloud processing

•Cloud-assisted security and privacy enhancement

V. POTENTIAL ENABLERS, CHALLENGES AND FUTURE

RESEARCH DIRECTIONS

In this section, we provide various potential enablers, key

challenges and potential future directions for the proposed

collaborative edge-cloud processing by categorizing them into

the following topics.

A. Coordination mechanisms between edge computing and

cloud computing

In the proposed collaborative edge-cloud architecture, com-

munication, computing and storage functionalities need to be

dynamically allocated among the edge-side units, cloud and

the things (devices/sensors) in order to handle the massive

IoT data in the real-time. Besides, there is a strong need to

have the coordinated management of cloud computing and

edge computing units in handling massive IoT connections

in a reliable and secured manner [8]. In other words, effective

coordination mechanisms between edge processing units and

cloud processing units are crucial in order to realize this

architecture effectively. In order to enable these interactions,

suitable interfaces between edge processing unit and the cloud

centre, among different edge processing units, and between

edge processing units and IoT devices/objects need to be

deﬁned.

Moreover, in the proposed coordinated platform, the cloud

center is assumed to be capable of dictating the edge com-

puting units about which parameters are important to monitor,

how frequent a parameter needs to be monitored and which

task should be processed to meet the real-time requirements of

end-users. In this regard, various aspects such as coordination

framework deﬁnition, parameters to be coordinated, control

signaling design from the cloud center, and load balancing at

the edge and cloud sides need to be investigated.

Some of the research challenges and future research direc-

tions under this research topic are provided below.

•Deﬁnition of edge-cloud interaction mechanisms: Vari-

ous aspects such as what should be the suitable interfaces

between cloud and edge, between cloud and things, be-

tween edge and things/devices and between different edge

computing units, which computing and communication

parameters will be involved in establishing relationship

between edge and cloud units, and how to design common

control signaling and management schemes in order to

control edge units from the cloud, need to be investigated

under the proposed framework.

•Resource mapping between edge and cloud comput-

ing: In order to enable effective collaborations between

edge and cloud units, it is important to investigate suitable

techniques to map edge-side computing/communication

resources with the cloud-side resources, and also suitable

strategies to share resources among multiple edge units

for handling live data anlytics.

•Load balancing among edge, cloud and things: It is

crucial to investigate which tasks can be handled at the

edge gateway and which tasks should be sent to the

cloud, which tasks can be handled at the things/devices,

in order to effectively optimize the available edge and

cloud resources in the proposed platform.

•QoS enhancement schemes: In the considered cloud-

edge coordinated platform, it is important to identify

the performance bottlenecks (communication bandwidth,

cache size, computing power, etc.) and then to develop

suitable techniques to minimize service response time and

to improve reliability in the case of network/link failures.

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

B. Big-data aware edge-cloud collaborative processing

One promising way of dealing with the big data in wireless

IoT networks is to understand the features of big data and

to incorporate this awareness in order to enhance the system

performance [15]. In contrast to multimedia contents, IoT data

has several peculiar features such as bursty nature, small data-

size and transiency, i.e., they expire in short time [31]. Besides,

in the proposed framework, only enriched data with the

meaningful information can be forwarded to the cloud instead

of sending all the raw data to the cloud. In this direction, it is

interesting to explore model-based data processing techniques

such as data compression, device cooperation, and distributed

coding/encoding by extracting certain features of IoT data such

as temporal and spatial correlations, and social relations. For

example, the data collected from IoT sensor nodes can be

compressed at the aggregator node before forwarding to the

cloud, and also only certain features of the raw data can be

extracted and sent to the cloud for some speciﬁc applications.

One of the main tasks of IoT sensors is to sense various

parameters of the environment under the practical constraints

such as low-cost, low-power and weak processing capability.

In the proposed architecture, it is crucial to optimize edge-

computing resources at the IoT gateway along with the com-

munication resources such as transmit power, bandwidth and

antennas in order to meet the desired performance metrics of

the proposed system such as latency, data rate and energy

efﬁciency. Also, in various IoT applications such as smart

health and smart car, it is important to acquire and process

contextual information such as location and speed to provide

meaningful information. In this regard, the cloud centre can

facilitate the optimization of edge-side processing under the

proposed framework. For example, sampling rate to be em-

ployed at the edge-node depends on the characteristics of

information (such as radio spectrum usage) to be acquired

from the environment, and higher sampling rate causes more

power consumption and also requires costly equipment. In

this example, the cloud centre can provide guidance to the

edge-nodes on the use of suitable sampling rate based on

its network-wide intelligence as well as history information

in order to reduce the power consumption. Besides, other

transmission and operating parameters at the edge-nodes such

as transmit power, modulation and coding can be adapted

based on the feedback provided by the cloud-centre.

Another key challenge in the proposed framework is how

to minimize the closed-loop (end-to-end) latency in order to

process IoT big data in the real-time. In practice, the closed-

loop latency of the considered system should be within the

data coherence time, i.e., lifetime of the data. By employing

suitable data prioritization techniques, the processing of the

prioritized data can be handled at the edge computing units

to provide faster response to the end-users. In addition, by

employing caching techniques at the edge computing units

in the proposed platform, content delivery efﬁciency can be

maximized and fronthaul/backhaul bandwidth overhead can

be signiﬁcantly reduced as in 5G wireless networks [24].

During the off-peak hours, popular contents can be pre-fetched

to the edge-side and during the peak periods, the delivery

phase has to only deal with the transmission of additional

contents requested from the users. The content popularity can

be predicted using the available historical big data at the cloud

side, and then a suitable cache replacement policy can be

employed on the basis of the predicted content popularity

distribution.

Based on the above discussion, we highlight some of the

research directions under this topic below.

•Energy-efﬁcient caching strategies: Various aspects

such as time-varying popularity estimation, popularity

modeling/prediction based on historical data, cache place-

ment, delivery and replacement strategies, different per-

formance tradeoffs such as data rate versus cache size

(memory), latency versus memory and energy versus

delay, cooperative and coded caching can be investigated

in the proposed framework.

•Edge-side data acquisition and processing techniques:

At the edge-side of the proposed platform, various cloud-

assisted data processing techniques such as data ﬁltering,

compression and feature extraction techniques, cooper-

ative sensing/monitoring/acquisition, distributed encod-

ing/decoding, combining and decision making schemes

can be investigated.

•Closed-loop latency minimization: In order to minimize

closed-loop latency in the proposed framework, suitable

device/node/task/data prioritization techniques can be in-

vestigated under various practical constraints such as

transmit power and backhaul/fronthaul bandwidth based

on different criteria such as residual energy, requirement,

emergency level, and expected delay.

•Adaptive learning/prediction algorithms: Several is-

sues such as how to make the best use of available pre-

diction algorithms (Neural network, artiﬁcial intelligence,

support vector machine, clustering, regression analysis,

etc.) in different application scenarios and how to adapt

algorithms based on time-varying situations are promising

future directions.

C. Task/data/computation/program ofﬂoading

It is understood that huge amount of IoT data cannot

be handled at the edge-side and need to be ofﬂoaded to

the cloud-side. Besides, resources at the edge side such as

transmit power, bandwidth and computing power need to allo-

cated efﬁciently to handle real-time applications at the edge-

side. Latency tolerant and large-scale tasks can be processed

efﬁciently at the cloud centre while it is advantageous to

process latency-sensitive tasks at the edge-side. The transfer of

computational load/data/task from the edge-side to the cloud

in the proposed collaborative edge-cloud framework can be

enabled by employing suitable task/data/computation/program

ofﬂoading techniques at the IoT edge gateway/aggregator. In

this regard, it is crucial to take effective decisions about

which data/task to ofﬂoad to cloud and which data/task to

be processed at the edge-side based on the guidance from the

cloud center.

By using suitable data prioritization techniques, critical

data (which needs real-time treatment) can be identiﬁed and

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

processed at the edge-side to provide faster response to the

end-users. The prioritized data can be later sent to the cloud

for further processing and storage for future usage. In this

regard, suitable data classiﬁcation, data prioritization, data/task

partitioning/scheduling, task/data ofﬂoading with backhaul

constraints (limited bandwidth, delay, power consumption)

and delay-based node prioritization need to be investigated

under the proposed framework. Besides, since there is a huge

amount of cost involved in renting cloudlets from the cloud

providers, it is essential to investigate the tradeoff between

service execution time and the cost required to use the cloud

resources [32].

Based on the above discussion, we provide some research

topics under this theme below.

•Task scheduling techniques: In the proposed framework,

suitable task scheduling techniques need to be investi-

gated for multiple dependent and interdependent tasks

under the constraints of cloud/edge computing resource

cost and service/workﬂow execution time deadline.

•Data ofﬂoading techniques: The investigation of suitable

data/task ofﬂoading techniques under backhaul and edge-

cloud resource constraints and decision making strategies

on which data to ofﬂoad to the cloud and which to process

at the edge-side are important future research directions.

•Code partitioning techniques: In order to maximize the

overall computational efﬁciency of the proposed frame-

work, it is crucial to investigate which parts of the code

to be run locally at the edge-side, and which parts should

be ofﬂoaded to the cloud and at what state of the program

considering practical constraints such as battery level,

delay constraint and channel state.

•Performance tradeoffs: Various performance tradeoffs

such as ofﬂoading gain versus energy, and cost of cloud

resources versus service execution time under the pro-

posed framework are interesting aspects to be investigated

in future works.

D. Adaptive optimization of computing, communications and

caching resources

In contrast to the traditional way of managing computing

and communication resources in a separate manner, future

wireless IoT networks require novel solutions towards the

adaptive optimization of computing, storage and communica-

tion resources at both the edge and cloud sides in order to

deal with the massive amount of heterogeneous data. Besides,

in the considered cloud-based IoT environment, the overall

system dynamics vary due to different causes such as device

movements, system parameter variations, and wireless channel

variations, thus making the designed system unstable over

the time. In this context, it is important to adapt the system

model and decisions/control actions to be taken based on the

global knowledge available at the cloud side and instantaneous

information collected at the edge-side. Moreover, objective

functions can be adapted based on the varying state of the

system as well as instantaneous requirements from the end-

users.

Besides, in order to facilitate real-time collaboration be-

tween the cloud computing and edge computing units, it is

crucial to optimize the involved backhaul/feedback links under

the transmission bandwidth constraint without compromising

the quality of the links. In this direction, suitable techniques

for designing low-latency ﬂexible backhaul links under the

constraints of end-user QoS requirements can be developed

by utilizing the concepts of software deﬁned wireless networks

[33]. Moreover, the interactions between various layers of pro-

tocol stacks need to be considered in devising end-end system

reliability. In this direction, various cross-layer techniques such

as modulation and coding adaptation at the physical layer,

collision free access mechanisms at the access layer and novel

routing algorithms at the network layer can be investigated for

the proposed architecture.

Based on the above discussion, we highlight some of the

interesting research topics below.

•Characterization of system resources: In the proposed

framework, it is important to identify all the involved

communication, caching and computing resources, and

then derive link/system capacity or any other suitable

metrics by considering all these resources into account

in order to provide the overall characterization of the

system.

•Adaptive optimization problems: Some examples in-

clude latency minimization under the constraints of

computational rate, transmit power minimization under

the constraints of computational rate, precoding ma-

trices/beamforming vector optimization under the con-

straints of computational rate, and the optimization of

processing power under the constraints of latency, band-

width and transmit power.

•Performance tradeoffs: In the considered framework,

several performance tradeoffs such as caching gain and

memory size/cache content size, computing and commu-

nication delays can be investigated.

•Joint optimization of system resources: Suitable joint

optimization solutions can be investigated under the

proposed framework in order to jointly optimize the

caching (cache size), computing resources (processor

speed, memory size, computational rate/power) and com-

munication resources (bandwidth, transmit power, energy,

latency etc.). Besides, suitable multi-objective optimiza-

tion solutions [34] can be developed in order to address

conﬂicting objectives in designing a wireless IoT net-

work.

E. Autonomous device collaborations in wireless IoT networks

Device to Device (D2D) communication has got impor-

tance in various practical scenarios and plays an important

role in the proposed edge-cloud coordinated framework in Fig.

3. The increasing context-aware applications require location

awareness and discovery functionalities and need to communi-

cate with other neighboring nodes. Also, D2D plays an impor-

tant role in enabling the sharing of resources such as contents,

applications, processing power and spectrum among spatially-

closed resource-constrained nodes in order to enhance the

overall system performance. In addition, D2D communication

is of vital importance to form emergency communication

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

network in disaster scenarios such as earthquake or hurricane

[35]. In such disaster scenarios, D2D communications has to

take care of various aspects such as unpredictability, limited

resources in disaster areas and dynamically varying environ-

ment [36].

In wireless IoT networks involving different types of

things/objects such as home appliances, body area sensors,

smart-phones and environmental monitoring sensors, a series

of communication is required to realize the process of creating,

collecting and sharing information among nearby devices [37].

Due to their heterogeneous capabilities, devices can help each

other with the help of suitable collaborative strategies among

them (for example, in a smart home, home audio, lighting

bulbs, and alarming systems can collaborate to indicate the

emergency level). Similar to human social network in con-

necting users via Internet, one device can be connected with

other devices based on its social relationships with them, thus

leading to the concept of social IoT [38]. By exploring the

social relations of the devices, device collaboration can be

initiated without human intervention.

The heterogeneity and the diversity of the connected smart

devices impose challenges in handling interoperability among

IoT devices. During the operation phase, there arise several

challenges such as how to maintain seamless connections

among the devices, how to attach a new device to the

network, how to rediscover a device in the network [39],

how to form a collaborative cluster, and which device to

disconnect in case of faulty conditions and security threats.

By grouping correlative devices under the same collaborative

cluster, efﬁcient management strategies can be employed at

the IoT gateway in order to control them, and also to enable

device collaborations. In this direction, the key issues to be

considered are to understand the social relations among the

devices [38], to classify devices based on a suitable basis, to

track the location of the devices/inhabitants [40], and to design

effective device collaborative strategies based on the inferred

information with the aim of enhancing the zero-conﬁguration

index of device interaction.

In the following, we highlight some of the interesting

research topics and issues under this theme.

•Deﬁnition of social relationships: Several aspects such

as how to assign social relations among IoT devices and

what kind of device social relations can be extracted from

human social relations are important to investigated under

the considered joint edge-cloud collaborative platform.

•Device attachment and discovery schemes: It is essen-

tial to develop suitable policies to admit a new device

to the existing IoT network and the ways to discover

a device in the shortest possible path to minimize the

latency as well as energy consumption.

•Techniques for device classiﬁcation and collaborative

cluster formation: Various research questions such as

what is the best classiﬁcation basis, how many devices

should be grouped in a network, how to support a large

number of heterogeneous group/cluster of devices, how to

synchronize different clusters, and how to choose a clus-

ter head dynamically in time varying channel conditions

need to be addressed in the proposed framework.

•Performance analysis: Balancing the tradeoff between

local device interactions and maintaining stable global

system state in the considered distributed system [12] is

one of the important aspects to be considered. Besides,

suitable incentive-based device participation approaches

can be investigated to enhance the participation of IoT

devices in sharing their communication, computing and

storage resources.

F. Data-driven event monitoring and event-driven service pro-

visioning in wireless IoT networks

In the considered cloud-edge collaborative framework,

various features of big data can be utilized in order to

monitor the events which may be periodic or random in

nature. Based on the global view of the network and historical

data, cloud can implement suitable prediction techniques to

forecast future events, and to provide this information to the

edge computing units in order to better utilize the network

resources. Moreover, in order to support device interactions

in IoT networks with the minimal human intervention, energy

efﬁciency, computational efﬁciency and spectral efﬁciency are

important issues to be considered. One of the promising

approaches to tackle these issues is event triggering, which

enables communication/computing action to take place only

when a particular event or a sequence of events happens

[41]. With this approach, resource utilization efﬁciency can

be signiﬁcantly improved since communication circuits can be

put into the sleep mode and the computing burden is reduced

during the non-occurrence of events.

The ﬁrst step in data-driven event monitoring is to deﬁne

a suitable behavior model which captures the normal behavior

of the considered IoT system. After a behavior model is iden-

tiﬁed, a suitable even detection technique can be employed.

The even detection techniques can be broadly categorized

into sample-based and stream-based [41]. The ﬁrst category

of techniques is based on individual samples to detect events

while the second category relies on the ﬂow of data samples.

Besides, the service provisioning of IoT is signiﬁcantly

different than from the traditional services which are mainly

targeted for human-machine interactions and the difference

mainly lies in the fact that IoT services need to deal with

the seamless interactions with the real-world [42]. The het-

erogeneous IoT environment requires distributed processing

of the large-scale sensing information collected by the IoT

nodes, which need to combined and shared among different

entities. In this regard, it is important to dynamically adapt the

services based on the instantaneous changes in the physical

environment.

We highlight some of the interesting research problems and

directions under this research topic below.

•Behavior modeling techniques: Several aspects such as

deﬁnition of suitable reference models, methods to cope

up with unpredictable system characteristics, exploitation

of differential states in the system state, and advanced

learning methods can be studied under the proposed

framework.

•Event detection techniques: Suitable sample and stream-

based mechanisms, and statistical approaches such as

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

eigenvalue distributions and PCA can be investigated for

the considered framework.

•Event handling methods: In order to handle events in

the proposed framework, various clustering techniques,

event classiﬁers, data-driven methods such as regres-

sion, expectation maximization and linear support vector

machines, inference methods such as fuzzy logic and

heuristic reasoning, and data fusion techniques can be

investigated [41].

•On-demand information sharing: Several aspects such

as resource sharing models, interactions with application

layer parameters, on-demand information dissemination

models, environmental awareness and information fusion

techniques require further research.

•Adaptation and validation of event triggering: In

practice, the events can be spontaneous, and may have re-

current patterns and are of dynamic nature. How to adapt

the operation of the network based on various states of the

event such as early sign, and intermediate state is another

key issue to be considered. After acquiring information

about the event, how to validate the credibility of that

information is another important challenge.

G. Software deﬁned networking and virtualization in wireless

IoT networks

In the existing networks, the main problems are vendor-

speciﬁc interfaces and software associated with hardware,

complex and expensive network operation, and the tight cou-

pling of data and control planes [26]. Besides, the network

cannot dynamically adapt based on the network conditions.

To address the aforementioned issues, the emerging concept

of SDN [26], [43] can be employed. Besides, virtualization

technology can be used to form a virtual network on the

top of the existing networks, which shields the user from

the underlying hardware and becomes adaptable to diverse

technologies and protocols [44].

However, in contrast to the traditional virtualized wired

networks, radio resource abstraction and isolation in wireless

networks is challenging due to time-varying channel, broad-

cast nature, mobility and heterogeneous access technologies

[27]. In heterogeneous IoT environment, dynamic resource

discovery and the sharing of available resources can enable

the creation of effective virtual networks. However, one of the

critical challenges is the efﬁcient management of the physical

resources allocated to virtual networks. How often resource

discovery and allocation need to be performed is also another

challenge to be addressed. Besides, mapping of the physical

resources to the logical resources in order to embed a virtual

network on the existing physical networks in an effective way

is another key aspect to be investigated.

The creation of a virtual network requires effective interac-

tion among the involved entities at different hierarchical levels

of the wireless IoT network [26]. This requires the need of

deﬁning suitable interfaces as well as proper control signalling

which can be adaptable among heterogeneous wireless access

technologies. For control signalling, mechanisms such as IP-

based signalling and dedicated channel assignment can be

investigated considering both network overload and delay.

Besides, naming the devices with logical identiﬁers, map-

ping between physical and logical addresses, slicing, device

attachment and dynamic routing of the device trafﬁc are other

important operations to be considered [26], [39]. Furthermore,

another key research aspect is to examine the possibility

of employing various levels of slicing such as spectrum-

level slicing, network-level slicing and ﬂow-level slicing in

the wireless IoT networks. In addition, investigating suitable

admission control policies in order to control the admission

of incoming users/devices while guaranteeing the QoS of the

existing users/devices is another aspect to be considered.

Based on the above discussion, some research directions

under this topic are provided below.

•Formation of virtualized edge-cloud collaborative net-

work: The design of a uniﬁed logical network in the pro-

posed framework include various aspects such as network

topology and interfaces deﬁnition, networks/devices dis-

covery/routing, mapping between physical and logical re-

sources, naming, assignment of unique logical identiﬁers

to the devices, and device attachment strategies/admission

control policies for future devices/nodes/users.

•Wireless virtualization technologies: It is an interesting

future research direction to investigate suitable radio

resource abstraction, isolation and slicing techniques un-

der various practical constraints such as time-varying

channel, mobility and heterogeneous access technologies

in the proposed edge-cloud collaborative framework.

•Adaptive conﬁguration of system parameters: Some

important research challenges in the proposed frame-

work are to dynamically conﬁgure a large number of

system parameters such as carrier frequency, bandwidth,

computing power, cache size and transmission power, to

estimate the time-varying wireless channels, i.e., loads,

and to dynamically characterize the communication links

in terms of stability, delay, rate loss, and collision. It

is interesting to investigate the application of SDN to

overcome these challenges.

•Control signalling design: Another key challenge is how

to design control signalling under the constraints of delay

and communication bandwidth overhead. Several ques-

tions such as whether it is better to provide a dedicated

radio channel or to employ a shared channel for control

signalling, how to support the scalability of the devices,

how to make signalling reliable in the case of device

mobility and device failure during D2D communication

need to be investigated in future works.

H. Security/privacy enhancement in wireless IoT networks

Another main challenge is how to provide secured con-

nections to the massive number of heterogeneous IoT devices

having different levels of processing capabilities. During the

processing of big data generated from massive IoT envi-

ronment, the security threat may arise in any of the data

processing phases including data acquisition, information ﬁl-

tering, data integration, representation, modeling, processing

and interpretation. In our proposed framework, the cloud

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

centre can guide the processing of edge nodes by utilizing

the available history information as well as the network-wide

intelligence towards achieving secured ﬂow of information in

the aforementioned phases.

The existing modular design practices for wireless commu-

nications are leading to vulnerable wireless IoT networks due

to the transparent air interface and weak security protection

mechanisms. Also, the decentralized characteristics of wireless

IoT networks, long sleeping windows of IoT sensors, and

the needs for collaborative communications to support D2D

communications over heterogeneous networks pose signiﬁcant

challenges to the traditional wireless security provisioning

techniques. For example, in IoT-based healthcare environment,

the delegation of both storage and computation to the untrusted

party can bring serious security and privacy issues and it

is crucial to maintain the security and privacy of the every

patient’s data [45].

Since the cloud centre has a global view of the network

as well as the history information, this knowledge can be

exploited in taking decision on which nodes are unreliable

and need to be disconnected from the network. In the IoT

environment involving heterogeneous sensors having different

levels of processing capabilities and also different mobility

levels, existing cloud computing security mechanisms like

sophisticated access control and encryption methods may not

be sufﬁcient to prevent illegitimate and unauthorized access to

the data. In this context, there is a strong need to investigate

suitable physical layer and cross-layer techniques such as dy-

namic link adaptation and adaptive medium access scheduling

to enhance the data ﬂow security in the proposed edge-cloud

coordinated platform.

In contrast to the traditional scenarios, there is a high

probability of privacy loss in IoT enabled systems due to

several factors such as location-based services, increased in-

teraction with smart devices, and less awareness on the user

side. In this regard, there is a strong need to investigate privacy

preserving mechanisms in IoT-enabled wireless networks [46].

Moreover, in the proposed cloud-edge collaborative platform,

the distributed edge units may be more vulnerable to attacks

since they do not have a global view of the network and

have less resources to protect themselves. In this regard, cloud

can assist edge-side units in implementing security measures

in practical scenarios with the help of its global intelligence

in identifying attacks/threats. Furthermore, in the proposed

platform, edge computing units can act as controllers and

aggregators for privacy sensitive data before sending data to

the cloud, and they may also act as the proxies of resource-

constrained IoT devices in order to handle their security

functions [12].

Based on the above discussion, the following speciﬁc

research directions can be considered under this theme.

•Security enhancement with joint edge-cloud process-

ing: Several research questions such as how to perform

cloud-side computations on the encrypted IoT data with-

out revealing any privacy/secrets to the cloud service

providers, how to ensure that data is not corrupted in

the edge processing units while in transit to the cloud

centre, and how to ofﬂoad tasks/programs from the edge

to cloud in a secured manner, need to be addressed.

•Access control and cooperative data aggregation

schemes: In the proposed platform, suitable cloud-

assisted encryption techniques, energy-efﬁcient data

scrambling techniques, and cooperative data aggregation

schemes can be investigated under the constraints of

latency and communication bandwidth. In the cooperative

schemes, various aspects such as how the nodes negotiate

for a shared key, and what are different ways in which

cloud can assist nodes in devising these cooperative poli-

cies under the constraints of communication bandwidth

and computational rate need to be investigated.

•Privacy preservation techniques: In the proposed cloud-

assisted platform, it is important to investigate suitable

privacy preserving data clustering, differential privacy

mechanisms and Pseudonymizing techniques in order to

preserve the privacy of the end-users.

•Trustworthiness of IoT systems: Various aspects such

as how to measure the trustworthiness of an IoT sen-

sor/aggregator node, how to identify that the source of

the data is a desired sensor device but not a robot or

a malicious device, need to be investigated under the

proposed framework.

VI. CONCLUSIONS

Cloud computing and edge computing are considered as

two emerging paradigms in handling the massive amount of

distributed data generated by IoT devices. However, these

paradigms have their own advantages and disadvantages.

Cloud computing provides a centralized pool of storage and

computing resources and has a global view of the network

but it is not suitable for applications demanding low latency,

real-time operation and high QoS. On the other hand, edge

computing is suitable for the applications which need real-time

treatment, mobility support, and location/context awareness

but does not usually have sufﬁcient computing and storage

resources. Taking these aspects into consideration, this paper

has proposed a novel framework of collaborative edge-cloud

processing for enabling live data analytics in wireless IoT

networks. The basic features, key enablers and the challenges

of big data analytics in wireless IoT networks have been

described and the main distinctions between cloud and edge

processing have been presented. Furthermore, potential key

enablers for the proposed collaborative edge-cloud computing

framework have been identiﬁed and the associated key chal-

lenges have been presented in order to foster future research

activities in this domain.

Finally, it is worthy to mention that the proposed edge-

cloud collaborative framework can be exploited as an im-

portant platform for wireless networks to achieve various

objectives such as dynamic spectrum management, energy-

efﬁcient caching and ofﬂoading, closed-loop latency mini-

mization, adaptive optimization of computing, communication

and caching resources, even-driven resource allocation and

security/privacy enhancement.

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

REFERENCES

[1] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and

M. Ayyash, “Internet of Things: A survey on enabling technologies,

protocols, and applications,” IEEE Communications Surveys Tutorials,

vol. 17, no. 4, pp. 2347–2376, Fourthquarter 2015.

[2] S. K. Sharma, T. E. Bogale, S. Chatzinotas, X. Wang, and L. B. Le,

“Physical layer aspects of wireless IoT,” in Proc. Int. Symp. on Wireless

Communication Systems (ISWCS), Sept. 2016, pp. 304–308.

[3] P. Fan, “Coping with the big data: Convergence of communications,

computing and storage,” China Communications, vol. 13, no. 9, pp.

203–207, Sept. 2016.

[4] H. Liu, Z. Chen, and L. Qian, “The three primary colors of mobile

systems,” IEEE Communications Magazine, vol. 54, no. 9, pp. 15–21,

Sept. 2016.

[5] S. Andreev, O. Galinina, A. Pyattaev, J. Hosek, P. Masek,

H. Yanikomeroglu, and Y. Koucheryavy, “Exploring synergy between

communications, caching, and computing in 5G-grade deployments,”

IEEE Communications Magazine, vol. 54, no. 8, pp. 60–69, August

2016.

[6] J. Tang and T. Q. S. Quek, “The role of cloud computing in content-

centric mobile networking,” IEEE Communications Magazine, vol. 54,

no. 8, pp. 52–59, Aug. 2016.

[7] P. Corcoran and S. K. Datta, “Mobile-edge computing and the Internet

of Things for consumers: Extending cloud computing and services to

the edge of the network,” IEEE Consumer Electronics Magazine, vol. 5,

no. 4, pp. 73–74, Oct. 2016.

[8] X. Masip-Bruin, E. Marn-Tordera, G. Tashakor, A. Jukan, and G. J. Ren,

“Foggy clouds and cloudy fogs: a real need for coordinated management

of fog-to-cloud computing systems,” IEEE Wireless Communications,

vol. 23, no. 5, pp. 120–128, Oct. 2016.

[9] C. Vallati, A. Virdis, E. Mingozzi, and G. Stea, “Mobile-edge com-

puting come home connecting things in future smart homes using

LTE device-to-device communications,” IEEE Consumer Electronics

Magazine, vol. 5, no. 4, pp. 77–83, 2016.

[10] M. Satyanarayanan, “The emergence of edge computing,” Computer,

vol. 50, no. 1, pp. 30–39, Jan. 2017.

[11] S. H. Park, O. Simeone, and S. S. Shitz, “Joint optimization of cloud and

edge processing for fog radio access networks,” IEEE Trans. Wireless

Communications, vol. 15, no. 11, pp. 7621–7632, Nov. 2016.

[12] M. Chiang and T. Zhang, “Fog and IoT: An overview of research

opportunities,” IEEE Internet of Things Journal, vol. PP, no. 99, pp.

1–1, 2016.

[13] S. Yin and O. Kaynak, “Big data for modern industry: Challenges and

trends [point of view],” Proceedings of the IEEE, vol. 103, no. 2, pp.

143–146, Feb. 2015.

[14] H. Hu, Y. Wen, T. S. Chua, and X. Li, “Toward scalable systems for big

data analytics: A technology tutorial,” IEEE Access, vol. 2, pp. 652–687,

2014.

[15] S. Bi, R. Zhang, Z. Ding, and S. Cui, “Wireless communications in the

era of big data,” IEEE Communications Magazine, vol. 53, no. 10, pp.

190–199, Oct. 2015.

[16] Y. He, F. R. Yu, N. Zhao, H. Yin, H. Yao, and R. C. Qiu, “Big data

analytics in mobile cellular networks,” IEEE Access, vol. 4, pp. 1985–

1996, 2016.

[17] H. Cai, B. Xu, L. Jiang, and A. V. Vasilakos, “IoT-based big data storage

systems in cloud computing: Perspectives and challenges,” IEEE Internet

of Things Journal, vol. PP, no. 99, pp. 1–1, 2016.

[18] D. Puthal, S. Nepal, R. Ranjan, and J. Chen, “Threats to networking

cloud and edge datacenters in the Internet of Things,” IEEE Cloud

Computing, vol. 3, no. 3, pp. 64–71, May 2016.

[19] J. A. Stankovic, “Research directions for the Internet of Things,” IEEE

Internet of Things Journal, vol. 1, no. 1, pp. 3–9, Feb. 2014.

[20] V. Cevher, S. Becker, and M. Schmidt, “Convex optimization for

big data: Scalable, randomized, and parallel algorithms for big data

analytics,” IEEE Signal Processing Magazine, vol. 31, no. 5, pp. 32–43,

Sept. 2014.

[21] R. Couillet and M. Debbah, Random Matrix Methods for Wireless

Communications, 1st ed. Cambridge University Press, 2011.

[22] S. K. Sharma, S. Chatzinotas, and B. Ottersten, “SNR estimation for

multi-dimensional cognitive receiver under correlated channel/noise,”

IEEE Trans. Wireless Communications, vol. 12, no. 12, pp. 6392–6405,

Dec. 2013.

[23] C. Zhang and R. C. Qiu, “Massive MIMO as a big data system: Random

matrix models and testbed,” IEEE Access, vol. 3, pp. 837–851, 2015.

[24] E. Zeydan, E. Bastug, M. Bennis, M. A. Kader, I. A. Karatepe, A. S. Er,

and M. Debbah, “Big data caching for networking: moving from cloud

to edge,” IEEE Communications Magazine, vol. 54, no. 9, pp. 36–42,

Sept. 2016.

[25] D. Mazza, D. Tarchi, and G. E. Corazza, “A cluster based computation

ofﬂoading technique for mobile cloud computing in smart cities,” in

Proc. IEEE Int. Conf. on Communications (ICC), May 2016, pp. 1–6.

[26] D. Kreutz, F. M. V. Ramos, P. E. Verssimo, C. E. Rothenberg, S. Azodol-

molky, and S. Uhlig, “Software-deﬁned networking: A comprehensive

survey,” Proceedings of the IEEE, vol. 103, no. 1, pp. 14–76, Jan. 2015.

[27] C. Liang and F. R. Yu, “Wireless network virtualization: A survey,

some research issues and challenges,” IEEE Communications Surveys

Tutorials, vol. 17, no. 1, pp. 358–380, Firstquarter 2015.

[28] R. Deng, R. Lu, C. Lai, T. H. Luan, and H. Liang, “Optimal workload

allocation in fog-cloud computing toward balanced delay and power

consumption,” IEEE Internet of Things Journal, vol. 3, no. 6, pp. 1171–

1181, Dec. 2016.

[29] S. Prathibha, B. Latha, and G. Sumathi, “Improving energy efﬁciency

of computing servers and communication fabric in cloud data centers,”

in Proc. Int. Conf. on Research in Computational Intelligence and

Communication Networks (ICRCICN), Sept. 2016, pp. 17–21.

[30] I. Park, D. Kim, and D. Har, “MAC achieving low latency and energy

efﬁciency in hierarchical M2M networks with clustered nodes,” IEEE

Sensors Journal, vol. 15, no. 3, pp. 1657–1661, Mar. 2015.

[31] S. Vural, P. Navaratnam, N. Wang, C. Wang, L. Dong, and R. Tafazolli,

“In-network caching of Internet-of-Things data,” in Proc. IEEE Int. Conf.

on Communications (ICC), June 2014, pp. 3185–3190.

[32] X.-Q. Pham and E.-N. Huh, “Towards task scheduling in a cloud-fog

computing system,” in Proc. Asia-Paciﬁc Network Operations and

Management Symposium (APNOMS), Oct 2016, pp. 1–4.

[33] J. Zhang, X. Zhang, and W. Wang, “Cache-enabled software deﬁned

heterogeneous networks for green and ﬂexible 5G networks,” IEEE

Access, vol. 4, pp. 3591–3604, 2016.

[34] E. Bjornson, E. A. Jorswieck, M. Debbah, and B. Ottersten, “Multiob-

jective signal processing optimization: The way to balance conﬂicting

metrics in 5G systems,” IEEE Signal Processing Magazine, vol. 31,

no. 6, pp. 14–23, Nov. 2014.

[35] M. N. Tehrani, M. Uysal, and H. Yanikomeroglu, “Device-to-device

communication in 5G cellular networks: challenges, solutions, and future

directions,” IEEE Communications Magazine, vol. 52, no. 5, pp. 86–92,

May 2014.

[36] J. Wang, Y. Wu, N. Yen, S. Guo, and Z. Cheng, “Big data analytics for

emergency communication networks: A survey,” IEEE Communications

Surveys Tutorials, vol. 18, no. 3, pp. 1758–1778, thirdquarter 2016.

[37] O. Bello and S. Zeadally, “Intelligent device-to-device communication

in the Internet of Things,” IEEE Systems Journal, vol. 10, no. 3, pp.

1172–1182, Sept. 2016.

[38] D. o. Kang, J. H. Choi, J. Y. Jung, K. Kang, and C. Bae, “SDIF: Social

device interaction framework for encounter and play in smart home

service,” IEEE Trans. Consumer Electronics, vol. 62, no. 1, pp. 85–93,

Feb. 2016.

[39] A. Papageorgiou, R. Bifulco, E. Kovacs, and H. J. Kolbe, “Dynamic

M2M device attachment and redirection in virtual home gateway en-

vironments,” in Proc. IEEE Int. Conf. on Communications (ICC), May

2016, pp. 1–6.

[40] M. Danancher, J. J. Lesage, and L. Litz, “Model-based location tracking

of an a priori unknown number of inhabitants in smart homes,” IEEE

Trans. Automation Science and Engineering, vol. 13, no. 2, pp. 1090–

1101, Apr. 2016.

[41] P. Kolios, C. Panayiotou, G. Ellinas, and M. Polycarpou, “Data-driven

event triggering for IoT applications,” IEEE Internet of Things Journal,

vol. PP, no. 99, pp. 1–1, 2016.

[42] S. Zhao, L. Yu, and B. Cheng, “An event-driven service provisioning

mechanism for IoT (Internet of Things) system interaction,” IEEE

Access, vol. 4, pp. 5038–5051, 2016.

[43] A. M. Akhtar, X. Wang, and L. Hanzo, “Synergistic spectrum sharing

in 5G HetNets: A harmonized SDN-enabled approach,” IEEE Commu-

nications Magazine, vol. 54, no. 1, pp. 40–47, Jan. 2016.

[44] N. Bizanis and F. A. Kuipers, “SDN and virtualization solutions for the

internet of things: A survey,” IEEE Access, vol. 4, pp. 5591–5606, 2016.

[45] J. Zhou, Z. Cao, X. Dong, and X. Lin, “Security and privacy in cloud-

assisted wireless wearable communications: Challenges, solutions, and

future directions,” IEEE Wireless Communications, vol. 22, no. 2, pp.

136–144, Apr. 2015.

[46] P. Porambage, M. Ylianttila, C. Schmitt, P. Kumar, A. Gurtov, and A. V.

Vasilakos, “The quest for privacy in the Internet of Things,” IEEE Cloud

Computing, vol. 3, no. 2, pp. 36–45, Mar. 2016.

http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Shree Krishna Sharma (S’12-M’15) received the

M.Sc. degree in information and communication

engineering from the Institute of Engineering, Pul-

chowk, Nepal; the M.A. degree in economics from

Tribhuvan University, Nepal; the M.Res. degree in

computing science from Staffordshire University,

Staffordshire, U.K.; and the Ph.D. degree in Wireless

Communications from University of Luxembourg,

Luxembourg in 2014. Dr. Sharma worked as a

Research Associate at Interdisciplinary Centre for

Security, Reliability and Trust (SnT), University of

Luxembourg for two years, where he was involved in EU FP7 CoRaSat

project, EU H2020 SANSA, ESA project ASPIM, as well as Luxembourgish

national projects Co2Sat, and SeMIGod. He is currently working as a

Postdoctoral Fellow at Western University, Canada. His research interests

include 5G and beyond wireless systems, Internet of Things (IoT), adaptive

optimization of distributed communication, computing and caching resources,

cognitive and cooperative communications, and interference mitigation and

resource allocation in heterogeneous wireless networks.

In the past, Dr. Sharma was involved with Kathmandu University, Dhu-

likhel, Nepal, as a Teaching Assistant, and he also worked as a Part-Time

Lecturer for eight engineering colleges in Nepal. He worked in Nepal Telecom

for more than four years as a Telecom Engineer in the ﬁeld of information

technology and telecommunication. He is the author of more than 70 technical

papers in refereed international journals, scientiﬁc books, and conferences.

He received an Indian Embassy Scholarship for his B.E. study, an Erasmus

Mundus Scholarship for his M. Res. study, and an AFR Ph.D. grant from

the National Research Fund (FNR) of Luxembourg. He received Best Paper

Award in CROWNCOM 2015 conference, and for his Ph.D. thesis, he received

“FNR award for outstanding PhD Thesis 2015” from FNR, Luxembourg.

He is a member of IEEE and has been serving as a reviewer for several

international journals and conferences; and also as a TPC member for a

number of international conferences including IEEE ICC, IEEE PIMRC, IEEE

Globecom, IEEE ISWCS and CROWNCOM.

Xianbin Wang (S’98-M’99-SM’06-F’17) is a Pro-

fessor and Canada Research Chair at Western Uni-

versity, Canada. He received his Ph.D. degree in

electrical and computer engineering from National

University of Singapore in 2001.

Prior to joining Western, he was with Commu-

nications Research Centre Canada (CRC) as a Re-

search Scientist/Senior Research Scientist between

July 2002 and Dec. 2007. From Jan. 2001 to July

2002, he was a system designer at STMicroelectron-

ics, where he was responsible for the system design

of DSL and Gigabit Ethernet chipsets. His current research interests include

5G technologies, signal processing for communications, adaptive wireless

systems, communications security, and localization technologies. Dr. Wang

has over 280 peer-reviewed journal and conference papers, in addition to 26

granted and pending patents and several standard contributions.

Dr. Wang is a Fellow of IEEE and an IEEE Distinguished Lecturer.

He has received many awards and recognition, including Canada Research

Chair, CRC President’s Excellence Award, Canadian Federal Government

Public Service Award, Ontario Early Researcher Award and ﬁve IEEE Best

Paper Awards. He currently serves as an Editor/Associate Editor for IEEE

Transactions on Communications, IEEE Transactions on Broadcasting, and

IEEE Transactions on Vehicular Technology and He was also an Associate

Editor for IEEE Transactions on Wireless Communications between 2007

and 2011, and IEEE Wireless Communications Letters between 2011 and

2016. Dr. Wang was involved in a number of IEEE conferences including

GLOBECOM, ICC, VTC, PIMRC, WCNC and CWIT, in different roles such

as symposium chair, tutorial instructor, track chair, session chair and TPC

co-chair.

Analyzing Threats and Attacks in Edge Data Analytics within IoT Environments

Article

Full-text available

Mar 2024

Citation: Mahadevappa, P.; Al-amri, R.; Alkawsi, G.; Alkahtani, A.A.; Alghenaim, M.F.; Alsamman, M. Analyzing Threats and Attacks in Edge Data Analytics within IoT Environments. IoT 2024, 5, 123-154. Abstract: Edge data analytics refers to processing near data sources at the edge of the network to reduce delays in data transmission and, consequently, enable real-time interactions. However, data analytics at the edge introduces numerous security risks that can impact the data being processed. Thus, safeguarding sensitive data from being exposed to illegitimate users is crucial to avoiding uncertainties and maintaining the overall quality of the service offered. Most existing edge security models have considered attacks during data analysis as an afterthought. In this paper, an overview of edge data analytics in healthcare, traffic management, and smart city use cases is provided, including the possible attacks and their impacts on edge data analytics. Further, existing models are investigated to understand how these attacks are handled and research gaps are identified. Finally, research directions to enhance data analytics at the edge are presented.

Development of an Intelligent Virtualization Platform Key Metrics Monitoring System: Collaborative Implementation with Self-Training and Bagging Algorithm

Article

Full-text available

Jun 2024
MOBILE NETW APPL

Ruey-Chyi Wu

In recent years, virtualization platforms have not only been used to integrate data from traditional application systems but have also actively collected Internet of Things (IoT) data from various network transmissions. To address the challenges of real-time monitoring for key metrics on virtualization platforms, this study proposes an optimal machine learning training model that combines semi-supervised Self-Training algorithms with supervised ensemble algorithms. In the application of semi-supervised training learning algorithms, this study utilizes a Self-Training learning algorithm to label a large number of unlabeled virtual machine operational states with a small amount of labeled data, laying the foundation for subsequent model construction. Subsequently, an ensemble learning classification algorithm is introduced to further validate and identify learning models suitable for generalization. Empirical evaluations show that the RandomForest algorithm serves as the optimal base estimator for Self-Training, while the Bagging algorithm is the optimal choice for ensemble learning. The synergy of these two achieves an accuracy exceeding 99%, enabling the model to accurately differentiate between various operational states such as normal operation, resource insufficiency, and faults. Finally, the integrated training model is deployed to a dashboard, displaying the real-time operational status of virtual machines through different colored lights. Simultaneously, operational status information is communicated to stakeholders through various media, further improving coordination, decision-making, and resource allocation issues on the virtualization platform. This study provides an efficient and feasible solution for monitoring and managing virtualization platforms.

A Study on Edge Computing Design Confrontation and It's Implementation in Internet of Things

Article

Full-text available

Mar 2024

The Edge computing paradigm has been increasingly well-liked in academic and business circles in recent years. By linking cloud computing facilities and services to end users, it acts as a major enabler for numerous upcoming technologies, including 5G, the Internet of Things (IoT), augmented reality, and vehicle-to-vehicle communications. Applications that are sensitive to delays can benefit from low latency, mobility, and location awareness offered by the Edge computing paradigm. Considerable investigation has been conducted in the field of edge computing, which is examined in light of recent advancements including fog computing, cloudlet, and mobile edge computing. resulting in providing researchers with more insight into the existing solutions and future applications. This article is meant to serve as a comprehensive survey of recent advancements in Edge computing highlighting the core applications. It also discusses the importance of Edge computing in real life scenarios where response time constitutes the fundamental requirement for many applications. The article concludes with identifying the requirements and discuss open research challenges in Edge computing.

Salient Applications and Obstacles in IOT Edge Computing

Article

May 2024

The Internet of Things (IoT) is a new technology that has emerged in recent years as a result of the exponential rise of linked gadgets. The IoT benefits from cloud computing. Applications to store the data and carry the calculations, that the enormous volume of data produced by these IoT devices may be managed and controlled. However, fulfilling the demands of numerous Internet of Things, real-time applications remains the main problem in cloud computing. On the other hand, Edge or Fog computing is a computing architecture that facilitates data management, processing, storing, and communication while providing a prompt response. By putting these functions closer to the end consumers, this is made possible. For many applications, edge and cloud computing are complementary technologies that work independently of one another. The overview of IoT and communication technologies and IoT protocols, as well as data transit in IoT. We look into cloud services for processing, storing, and analysing the data produced by Internet of Things devices.

IoT Data Stream Handling, Analysis, Communication and Security Issues: A Systematic Survey

Article

Full-text available

May 2024
WIRELESS PERS COMMUN

Massive data management or data stream handling has become a prominent study topic as a result of recent breakthroughs in the Internet of Things (IoT). With the developments in big data technologies and their applications in the IoT realm, the analyzing, processing,, and handling streams of data and information are attaining the vast attention of researchers. In IoT, continuous information generation, handling and secure communication are achieved through innumerable devices. This information is enormous and time-varying. In all IoT applications processing, analyzing, and handling such information are extremely critical, as the ceaseless flow of information and time-sensitive decisions may make it inefficacious to process after storage. Therefore, processing such massive or streams of information must be executed in real-time. A better understanding of the stream data handling, analysis and communication aspect can help in effective data management. This review work provides the vitality of secure data stream handling and analysis in IoT applications. It explains the critical features and requirements of IoT data for the analysis and security concerns in industrial applications like smart city, smart agriculture and smart transportation system. It explores various studies emphasizing massive data management and the complexities involved with it. This article further investigates the different models and techniques utilized for handling IoT stream data in existing research works and enumerates their merits in stream data performance management. The review work also describes the technical challenges and research gap existing in-stream data handling and crucial further developments needed in this arena for effective management of IoT stream data.

IoMT Landscape: Navigating Current Challenges and Pioneering Future Research Trends

Preprint

Full-text available

May 2024

Smart electronic devices and telemedicine are widely used in humans' daily lives. This provides remote treatment of patients using information and communications technology. New telemedicine technologies, such as the Internet of Medical Things (IoMT), make it easier for medical and computing devices to communicate regularly and effectively. Critical motivations for adapting the IoMT are reduced cost, increased quality of life, and timely medical intervention. IoMT is significant because it enables continuous, real-time patient monitoring during routine everyday activities using a variety of wearables and sensors. With big data, the IoMT technology makes excellent use of Artificial Intelligence (AI) to support disease detection and health condition prediction, alerting patients and healthcare providers. Many research studies have been conducted to explore various aspects of IoMT and its applications in the real world. However, it is challenging to comprehend all the techniques and solutions proposed by the research community. Therefore, this survey sheds light on some crucial aspects of IoMT technologies and explores the potential research gaps and directions the research community could tackle. The survey examines and discusses the characteristics of IoMT standards, protocols, and types. It then delves into the layers of IoMT and distinguishes them into Fog and edge. The studies published under each type were explored, and the limitations of these works were highlighted. The research gaps and directions on IoMT approaches and technologies were also highlighted. With such findings and research directions, further research endeavors could be carried out to address the issues and existing limitations in the IoMT.

DCNN and LSTM based Plant Leaf Disease Detection Model For Edge IoT Device

Conference Paper

Feb 2024

Real-Time Event Detection and Predictive Analytics Using IoT and Deep Learning

Chapter

Mar 2024

The internet of things (IoT) has led to an explosive increase in connected devices, generating a massive volume of data. Real-time analytics in IoT systems is crucial for timely decision-making, enhancing system efficiency and reliability. This involves processing discrete IoT data series within a bounded completion time, providing services like data classification, pattern analysis, and tendency prediction. However, the continuous and heterogeneous generation of IoT data poses significant technical challenges. Designing IoT systems to handle this data in a timely manner becomes critical. This chapter comprehensively explores real-time data analytics in IoT systems, elucidating its characteristics and analysing suitable architectures. A survey of existing applications highlights system design perspectives and performance shortcomings. Lastly, challenges in applying real-time analytics in IoT systems are identified, paving the way for future research directions.

Walkthrough into Cloud Security

Article

Full-text available

Dec 2020

With the recent advancements and migration of organizations to the cloud environment, there is also an increase in the vulnerabilities and threats in the cloud. A wide extent of attacks that are pertinent to a PC and the data in transmission likewise applies to cloud based organizations. And nowadays more are adopting a multi-cloud strategy and its security becomes a more and more complex task. A lot of security techniques and algorithms have been proposed on the security in the cloud. This paper describes the various techniques adopted in the security on the cloud. The walkthrough technique builds up a primary corpus of information upon which a more detailed analysis and meanings can be built later on. It also serves as a foundation for further security centered research. Also a two-phase encryption method has been proposed to provide maximum security.

Knowledge-Driven Approach for Quality Assessment of HAR Data Sets: An Automated Tool

Conference Paper

Aug 2023

Foggy clouds and cloudy fogs: A real need for coordinated management of fog-to-cloud computing systems

Article

Full-text available

Oct 2016

The recent advances in cloud services technology are fueling a plethora of information technology innovation, including networking, storage, and computing. Today, various flavors have evolved of IoT, cloud computing, and so-called fog computing, a concept referring to capabilities of edge devices and users’ clients to compute, store, and exchange data among each other and with the cloud. Although the rapid pace of this evolution was not easily foreseeable, today each piece of it facilitates and enables the deployment of what we commonly refer to as a smart scenario, including smart cities, smart transportation, and smart homes. As most current cloud, fog, and network services run simultaneously in each scenario, we observe that we are at the dawn of what may be the next big step in the cloud computing and networking evolution, whereby services might be executed at the network edge, both in parallel and in a coordinated fashion, as well as supported by the unstoppable technology evolution. As edge devices become richer in functionality and smarter, embedding capacities such as storage or processing, as well as new functionalities, such as decision making, data collection, forwarding, and sharing, a real need is emerging for coordinated management of fog-to-cloud (F2C) computing systems. This article introduces a layered F2C architecture, its benefits and strengths, as well as the arising open and research challenges, making the case for the real need for their coordinated management. Our architecture, the illustrative use case presented, and a comparative performance analysis, albeit conceptual, all clearly show the way forward toward a new IoT scenario with a set of existing and unforeseen services provided on highly distributed and dynamic compute, storage, and networking resources, bringing together heterogeneous and commodity edge devices, emerging fogs, as well as conventional clouds. Introduction: The Scenario

Physical layer aspects of wireless IoT

Conference Paper

Full-text available

Sep 2016

An Event-Driven Service Provisioning Mechanism for IoT (Internet of Things) System Interaction

Article

Full-text available

Sep 2016

The heterogeneity, large scale, and resource constraint of IoT environment raise some issues which hinder its development. We focus on two issues among them: (1) most of the existing IoT applications are ”silos”, that is, WSAN (Wireless Sensor and Actuator Network) resources and applications are tightly coupled, applications cannot share and reuse resources and interact with each other; (2) how to efficiently disseminate the sensing information among the information providers and consumers, and rapidly respond to changes in the physical world. This paper proposes a MM (multilevel and multidimensional) model based service provisioning platform which can access large-scale heterogeneous resources and expose their capabilities as light-weighted services. Moreover, it presents a unified message space to facilitate the on-demand dissemination and sharing of sensing information in distributed IoT environment. The platform supports applications to share and reuse resources and provides the basic infrastructure for the IoT application pattern: inner-domain autonomy and inter-domain coordination.

Improving energy efficiency of computing servers and communication fabric in cloud data centers

Conference Paper

Sep 2016

The Emergence of Edge Computing

Article

Jan 2017

Mahadev Satyanarayanan

Industry investment and research interest in edge computing, in which computing and storage nodes are placed at the Internet's edge in close proximity to mobile devices or sensors, have grown dramatically in recent years. This emerging technology promises to deliver highly responsive cloud services for mobile computing, scalability and privacy-policy enforcement for the Internet of Things, and the ability to mask transient cloud outages. The web extra at www.youtube.com/playlist?list=PLmrZVvFtthdP3fwHPy-4d61oDvQY-RBgS includes a five-video playlist demonstrating proof-of-concept implementations for three tasks: assembling 2D Lego models, freehand sketching, and playing Ping-Pong.

IoT-Based Big Data Storage Systems in Cloud Computing: Perspectives and Challenges

Article

Jan 2016

Internet of Things (IoT) related applications have emerged as an important field for both engineers and researchers, reflecting the magnitude and impact of data-related problems to be solved in contemporary business organizations especially in cloud computing. This paper first provides a functional framework that identifies the acquisition, management, processing and mining areas of IoT big data, and several associated technical modules are defined and described in terms of their key characteristics and capabilities. Then current research in IoT application is analyzed, moreover, the challenges and opportunities associated with IoT big data research are identified. We also report a study of critical IoT application publications and research topics based on related academic and industry publications. Finally, some open issues and some typical examples are given under the proposed IoT-related research framework.

Towards task scheduling in a cloud-fog computing system

Conference Paper

Oct 2016

Coping with the Big Data: Convergence of Communications, Computing and Storage

Article

Sep 2016
CHINA COMMUN

Pingzhi Fan

To cope with the impact of big data, it is proposed to integrate the traditionally individual computing, communications and storage systems, which are getting inevitably converged. An effective information system capacity is introduced and discussed, aimed at excavating potentials of information system under a new paradigm with more degrees of freedom.

Mobile-Edge Computing and the Internet of Things for Consumers: Extending cloud computing and services to the edge of the network

Article

Oct 2016

Current activities in the Internet of Things (IoT) are focused on architectures, protocols, and networking for the efficient interconnection of heterogeneous things, infrastructure deployment, and creation of value-added services. The majority of the IoT products, services, and platforms are supported by cloud-computing platforms. With the IoT being a multidisciplinary ecosystem, it is now being utilized in connection with scenarios demanding real-time data processing and feedback, for example, connected and autonomous vehicles scenarios. Cloud platforms are not suitable for scenarios involving real-time operation, low latency requirements, and high quality of service (QoS). Recently, mobile-edge computing (MEC) has gained momentum from the industry to address the mentioned requirements. MEC is a novel paradigm that extends cloud-computing capabilities and services to the edge of the network. Due to dense geographical distribution, proximity to consumers, support for high mobility, and open platform, MEC can support applications and services with reduced latency and improved QoS. Thus, MEC is becoming an important enabler of consumer-centric IoT applications and services that demand real-time operations. The OpenFog Consortium and standards development organizations like ETSI have also recognized the benefits the IoT and MEC can bring to consumers. Potential applications for MEC-enabled IoT include smart mobility, connected vehicles, emergency response, smart cities, content distribution, and location-based services.

Mobile-Edge Computing Come Home! Connecting things in future smart homes using LTE device-to-device communications

Article

Oct 2016

Future 5G cellular networks are expected to play a major role in supporting the future Internet of Things (IoT), due to their ubiquitous coverage, plug-and-play configuration and embedded security. Besides connectivity, however, IoT will need computation and storage in proximity of sensors and actuators to support time-critical and opportunistic applications. To this aim, the introduction of Mobile Edge Computing (MEC) is currently under standardization as a novel paradigm to enable hosting of applications straight into the network. In this work we analyze solutions to bring MEC functionalities as close as possible to end users and smart objects. First, a smart-home system is designed as a reference small-scale IoT system to derive network require-ments; then alternative network configurations to support such requirements are analyzed to highlight their pros and cons. In particular, we show how LTE Device-to-Device (D2D) operation mode can be exploited to guarantee proximity communication with reduced costs. Finally, the expected benefits for operators are assessed via simulation.

Live Data Analytics With Collaborative Edge and Cloud Processing in Wireless IoT Networks

Abstract and Figures

Recommended publications

Cloud and Fog Computing for the IoT

Secure and Access Control Data Monitoring in Vehicular Ad Hoc Network

Data Processing Delay Optimization in Mobile Edge Computing

Real-Time Data Analytics in Internet of Things Systems

Emerging Edge Computing Technologies for Distributed IoT Systems