2.

Literature Survey

2.1 Background:

In the past decade, cloud computing has emerged as a popular platform for processing user
requests in cloud data centres [107]. Simultaneously, the demand for Internet of Things (IoT)
connected devices and applications, such as smartphones, gadgets, and Google Glass, has
significantly increased [132]. This surge in demand has led to challenges in optimizing
Quality of Service (QoS) parameters like bandwidth, privacy, latency, storage, response time,
security, and computing [36, 170]. Fog computing, a paradigm that integrates IoT concepts,
has shown promise in addressing these challenges by enabling communication and
interaction among nodes and devices [190]. Several frameworks and architectures for fog
computing have been proposed in the literature [49, 117, 177], targeting various IoT
applications. However, existing review papers have not presented a taxonomy of fog
computing frameworks. Hence, there is a need to design a taxonomy that categorizes fog
computing frameworks and identifies research gaps. This article presents a systematic
literature review of existing works, focusing on frameworks and architectures related to fog
computing. The findings of this review will assist researchers in advancing their studies in the
field of fog computing.

Context With the increasing number of IoT devices, mobile internet, and interconnected
objects, a massive amount of data is being generated [38]. It is estimated that by 2022, around
60-70 billion devices will be connected to the internet, resulting in a significant data transfer
[152]. Traditional computing models like cloud and distributed computing face challenges in
handling this scale of data [60]. Certain applications require quick response and mobility
support, such as emergency response in smart healthcare, traffic light systems, and smart
grids [172]. To address issues like high bandwidth, ultra-high latency, privacy-sensitive
applications, and geographic dispersion, a computing paradigm is needed to assist cloud
computing and process requests from connected IoT devices with minimal latency and
response time [11].

The concept of fog computing was introduced by Cisco in 2012 to address the challenges of
IoT applications in traditional cloud computing models [30]. It has gained support from the
OpenFog Consortium, which has published various white papers on the subject. The
consortium defines fog computing as "A horizontal system-level architecture that allocates
storage, computing, networking functions, and controls the nearest connected devices" [129].
Fog computing acts as a bridge between IoT nodes/devices and the cloud, enabling storage,
data management, networking, and computing on network nodes [59, 108, 109]. It reduces
the burden on traditional cloud computing data centers [96, 120] and operates on the concept
of decentralized computing to enhance computational speed. Currently, fog computing plays
a significant role in various domains such as agriculture, healthcare, construction, industry,
smart cities, smart homes, energy grids, transportation, environmental monitoring, and
climate change monitoring [22, 79, 131, 156].

Related Surveys In this section, we discuss previous surveys conducted on frameworks and
architectures based on fog computing. To the best of our knowledge, no systematic review
focusing on the taxonomy of fog computing frameworks exists in the literature. However,
several authors have independently proposed frameworks and architectures for fog computing
in various IoT application domains [169]. Generally, frameworks are sets of rules and tools
used to create and produce something [44]. They provide a pre-built composition or structure
specifically designed to extend specialized or general-purpose functionality [102].
Frameworks establish standard and specific guidelines to assist in the creation and extension
of valuable services.

Habibi et al. [69] developed a taxonomy covering security, software, applications, systems,
networking, and resource management in fog computing. They conducted a comprehensive
study on architectural perspectives, describing the logical and physical components of the fog
computing environment. The authors compared and evaluated proposed architectures based
on various criteria. However, this review only focused on the architectural aspect of fog
computing.

Margariti et al. [113] presented a survey of simulation tools and modeling methods to
examine system costs in terms of configuration, network, and applications. They provided an
overview of computing simulation tools, highlighting their characteristics, limitations, and
efficiency.

Brogi et al. [33] conducted a review of methodologies for solving the service placement
problem in fog computing. They provided a comprehensive summary of existing algorithms
and open-source prototypes. The authors focused on infrastructure characteristics and
optimized metrics.
Markus and Kertesz [114] focused on the classification of software quality. They analyzed
source code and its features in various simulators.

Yousefpour et al. [190] conducted a survey and proposed a taxonomy for fog computing.
They discussed previous papers that focused on frameworks and architectures of fog
computing.

Hou et al. [75] emphasized frameworks related to vehicular fog computing, including
conventional fog devices, path computing, enlisted edge computing, and fog nodes as IoT
hubs. They presented an overview of the resource model, generic architecture, and resource
allocation framework for the fog computing model.

Fernando et al. [49] discussed the central elements of fog computing frameworks and
architectures, highlighting opportunities for further research. They investigated and compared
industry-related frameworks, considering parameters such as low latency, dynamicity,
scalability, context-aware offloading, and heterogeneity that support the fog computing
framework.

Mouradian et al. [117] conducted a comprehensive survey on fog computing and examined
state-of-the-art research issues in the field.

Fog nodes specifically cloudlets, networking devices, base stations, vehicles and their
parameters are specified subsequently in the literature.

SLOs
Fog Nodal Provisioning Applicable
Work (ServiceiLevel
Nodes Collaboration Metrics Network
iObjectives)
Leeietial Network
Servers Master-slave Data (flow) IoT
[16] Management
Aazamie
Resource
tiial. Servers – Context (user) IoT
Management
[17]
Jalaliieti CDN (content
Time Power
al. Servers PeeritoiPeer distribution
(computing) Management
[18] network)
Zhuietial Servers - Context (user) Application CDN
[19] Management

Zengieti
Time Resource
al. Servers PeeritoiPeer IoT
(computation) Management
[20]
Hongieti
Networki Application
al. PeeritoiPeer Data (size) IoT
Devices Management
[21]
PLC
Nazmud
Networki Data (Poweriline
een MasteriSlave Data (size)
Devices Management communicatio
[22]
n)
Aazamie
Networki Network
tial. – Data (size) IoT
Devices Management
[23]
Ciraniiet
Networkid Context Network
ial. – IoT
evices (application) Management
[24]
Dsouzai
Context Network
et Cloudlets PeeritoiPeer IoT
(application) Management
[25]
Cardelli
Context Application
niiet Cloudlets Cluster IoT
(application) Management
[26]
RAN
Yanietial Base Application
Cluster Context (user) (RadioiAccess
[27] Stations Management
Network)
Cost
Guietial. Base (deployment, Resource
PeeritoiPeer IoT
[28] Stations communicatio Management
n)
Truongie Base Context(applic Network Vehicular
Master-slave
tial. Stations ation) Management network
[29]
Oueisieti
Base Latency Mobile
al. Cluster Time(deadline)
Stations Management network
[30]
Houietia
Resource
l. Vehicles Cluster Context(user) Vehicular
Management
[31]
Yeietial.
Time Application
i Vehicles – Vehicular
(deadline) Management
[32]
Oueisieti
Base Resource Mobile
al. Cluster Data(flow)
Stations Management network
[33]
Faruquei
Network Energy Power
et Cluster IoT
Devices consumption Management
[34]
Shiietial. Networki Context Application Mobile
PeeritoiPeer
[35] Devices (application) Management network
Giangiet
Networki Application
ial. PeeritoiPeer Data (flow) IoT
Devices Management
[36]
Intharaw
Time Latency Mobile
ijitr Servers –
(computation) Management network
[37]
Pengieti
Base Network
al. PeeritoiPeer Data (size) RAN
Stations Management
[38]
Hassanie Cluster
Network Application Mobile
tial. (execution,icom Cost
Devices Management network
[39] munication)
Zhangiet LRPON
Costi(deploym
ial. Cloudlets PeeritoiPeer Cost (Longireachip
ent)
[40] assiveiopticali
 network)

Dengieti
Network Application Mobile
al. PeeritoiPeer Data (Size)
Devices Management network
[41]
Doietial. Network Energy
– CiiO2imanagement CDN
[42] Devices consumption
Aazamie
Resource
tial. Servers – Context (user) IoT
Management
[43]
Dattaieti
Network Data Vehicular
al. PeeritoiPeer Context(user)
Devices Management network
[44]
Aazamie
Resource
tial. Servers – Context (user) IoT
Management
[45]
Gazisieti
Network Context(applic Resource
al. – IoT
Devices ation) Management
[46]

Table 2.1: Comparison of Fog nodes

2.2 Nodal Collaboration:
2.2.1 Cluster:
Fog nodes are capable of forming cluster among themselves as a consequence they can
sustain a collaborative execution environment. Their homogeneity [26] and location [31] can
be a basis for the formation of Fog nodes. Higher priority can also be given to computational
load balancing [33] and functional sub-system development [34] during the formation of
cluster.
The collaboration is potent in harnessing the capabilities of several Fog nodes concurrently.
However, it is arduous to make static clusters scalable in run-time. Besides, the dynamic
formation of clusters chiefly depends on the attainability and the current load of Fog nodes.
In either of the cases, a crucial role is played by networking overhead.
2.2.2 Peer to Peer:
Hierarchical [21] and flat order [35] can be a classification for Peer to Peer. The
collaboration can also be categorized as home, local, non-local etc. [18] based on proximity.
Processed output of one node is capable of appearing as input to some other node [36] as well
as virtual computing instances also get shared between the nodes [25] through P2P
collaboration.
2.2.3 Master-Slave:
A master Fog node controls the processing load, functionalities, resource management, data
flow, etc. [16] of underlying slave nodes generally.
Master-slave approach collectively with cluster and P2P based nodal interactions are capable
of forming a hybrid collaborative network [22,29]. Regardless, the master and the slave Fog
nodes require high bandwidth to communicate with each other owing to this type of
functional decomposition.
2.3 Resource/Service Provisioning Metrics:
Resource/service provisioning metrics are mentioned as follows.
2.3.1 Time:
• Computation time:
It can be adjusted per the existing load and substantially depends on the resource
configuration where the task has been scheduled [20], or the application is running. It also
helps in the identification of active and idle periods of various applications which are notable
determinants of resource and power management [18].
• Communication time:
It is defined as the networking delay to trade data elements. End device/sensors to Fog nodes
[37] and Fog nodes to Fog nodes [20] can be the categorizations of computation time. The
communication time reflects the network context, which assists the selection of suitable Fog
nodes for executing tasks.
2.3.2 Deadline:
It refers to the paramount delay that can be tolerated by a system in service delivery.
Deadline satisfied task completion is reckoned an essential parameter for measuring QoS of
the system [30,32]. Service delivery deadline plays a crucial part in portraying latency-
sensitive and latency tolerant applications.
2.3.3 Data:
• Data size:
It refers to the measurement of data that has to be Fog computing processes. Numerous works
have discussed data size in regards to computational space provisions of requests [35].
Besides, bulk data collected from distributed devices/sensors incorporate features of Big Data
[22]. It can be efficacious to provision services and resources according to the data load in the
above case. Furthermore, data size is crucial in deciding either local or remote processing of
the computational tasks [38].
• Data flow:
It describes the attributes of data transmission. Data flow can be event-driven [16] or real-
time [36] and factors into associated service and asset provisioning significantly. Besides,
dynamic load balancing is fostered by a sudden change in data flow[33].
2.3.3 Cost:
• Networking cost:
Associated expenses and bandwidth requirements are directly related to Fog computing
environment. In various works, internodal data sharing cost and cost of uploading from end
devices/sensors are fundamentals of networking cost [28]. Whereas, in other literature,
networking cost [39] is specified as network latency experienced due to bandwidth issues
[39].
• Deployment cost:
It is linked with the expenses associated with infrastructure placement in Fog computing
environment. Some papers have reasoned that cost-effective infrastructure deployment is
approving of productive service and resource provisioning. Placement of nodes in the
network [40] and the creation of virtual instances [28] can be used to discuss infrastructure
deployment cost.
• Execution cost:
It is defined as the computational cost of Fog nodes while running applications or processing
tasks. Despite wide usage of execution cost in billing and resource provisioning of other
computing paradigms, the metric has displayed minute usage in Fog computing literature.
These works have calculated the total execution cost regarding per unit time resource usage
cost and task completion time [39].
2.3.4 Energy consumption & Carbon Footprints:
The expenditure of energy in the energy-latency trade-off [41] and home-based Fog
computing environment [34] in distinct phases of Fog-Cloud interaction have been
emphasized prominently in literature. A paper mentions the carbon emission rate in terms of
per unit energy usage in distinct nodes has been taken note of for resource provisioning
purposes [42]. The energy aspects of end components, for example, energy characteristics of
communication medium, residual battery lifetime can also be researched in provisioning Fog
resources as end sensors/devices are energy-constrained.
2.4 Service Level Objectives:
2.4.1 Latency Management:
Latency management resists the ultimate service delivery time from exceeding an accepted
threshold. Applications QoS requirement or latency of a service request can be an appropriate
threshold. In some literature, the efficient initiation of nodal collaboration for the execution
of tasks by collaborated nodes within the enforced latency constraints [30] is highlighted. In
another paper, low-latency Fog network architecture has been suggested for handling latency
[37]. Besides, in another paper, the computation task distribution between the client and Fog
nodes have been conducted to reduce the communication latency of service requests and total
computation [20]. The essential purpose of this work is to select the node which contributes
to the lowest latency in service delivery.
2.4.2 Cost Management:
Operating Expenses (OPEX) and Capital Expenses (CAPEX) can be used to discuss cost
management. Deployment of cost of distributed Fog nodes and their associated networks is
the main contributor CAPEX. Optimized number and appropriate placement of Fog nodes are
imperative in lessening the CAPEX in such cases. A Fog computing network architecture that
optimizes the numbers and places of Fog node deployment has been proposed [40] in regards
to this issue.
For launching VMs Fog nodes have been contemplated as virtual platforms in [28]. In these
VMs execution of data processing operations is not free of cost. It is manageable to minimize
OPEX in Fog computing by exploiting the cost-diversity in different Fog nodes as the cost of
data processing operations can be varied from distinct providers. In turn, a solution to find an
apt set of Fog nodes for hosting VMs has been proposed in the paper.
2.4.3 Network Management:
• Network congestion:
It generally arises due to escalating overhead on the network. Simultaneous interactions of
end constituents with Cloud data centres and dispersion of IoT end sensor/devices across the
edge can inflate overhead on the core network. This results in network congestion which may
degrade the performance of the system. A layered architecture of Fog node that offers local
processing of the service requests has been proposed taking cognizance of this into account.
Outcome of this is that even while receiving bulk service requests, Clouds get a brief version
of the requests.
• Virtualization:
SDN is considered as one of the key backers of the virtualized network and offers support for
NFV. SDN is a networking approach that enacts in software on separate servers after
decoupling the control plane from networking gear. The architectural concept that virtualizes
of traditional networking functions firewalling, (network address translation (NAT), caching,
domain name service (DNS), intrusion detection, etc.) is called NFV. Several works [16, 29,
38] have suggested new network structures that enable SDN and NFV, this is due to the
broad range of services provided by SDN and NFV making them crucial in a Cloud-based
environment.
2.4.4 Computation Management:
• Resource estimation:
Helps in the allocation of computational resources per some policies for appropriate resource
allocation, achieving desired QoS and imposing accurate service price. Existing literature has
developed resource estimation policies concerning experienced QoE, user characteristics,
features of service access devices, etc. [17, 43, 45].
• Workload allocation:
Work Allocation is for ensuring a balanced load on various components. To equalize the
computational load on Fog nodes/client devices, scheduling based workload allocation policy
has been introduced in [20].Which results in overhead on both parts becoming economical
and amending QoE. In another work [41] to balance delay and power expenditure, a
workload allocation framework has been proposed.
2.4.5 Application Management:
• Programming platform:
To develop, compile and execute applications programming platforms offer essential
components such as run-time environment, interfaces, libraries, etc. "Mobile Fog" [21] has
been created to overcome the issue of proper programming support (due to the dynamic
nature of Fog computing). It offers support for developing extensive applications over
heterogeneous-distributed devices by simplifying abstraction of programming models. A
programming platform based on distributed data flow approach has also been designed in
[36] for application development, besides coordinating resources during applications
execution.
• Scaling:
The adjustment capability to retain service quality despite unpredictable events and the
conception of application users. They are also applicable to users service access and
application scheduling. Architecture of a QoS aware self-adaptive scheduler to assist scalable
scheduling of data stream applications has been suggested recently [26]. This scheduler is
capable of scaling applications with the increase of resources and users while not requiring
global information about the environment. Furthermore, due to the self-adaptive capability of
the scheduler, placement of applications in dispersed fashion and automatic reconfiguration
of the resources become more manageable. An adaptive procedure for user’s service access
mode selection based on distance, location and QoS requirements of the service reaching
entities has also been proposed [27].
2.4.6 Data Management:
Data management is an imperative SLO necessary for efficient performance. Various
research papers have discussed it from differing perspectives. In [23, 44] resource allocation
for data pre-processing and initiation of proper data analytic services are focused on data
management policy. Also, low-latency aggregation of data coming from distributed end
devices/sensors can also be considered for efficient data management [22]. Besides, storage
augmentation in Fog computing for preserving data of end entities can be highly effective as
the storage capacity of end devices/sensors are poor. Accordingly, in [39] storage expansion
in Fog computing for mobile devices has also been presented, besides application
administration, as an elemental part of data management.
2.4.7 Power Management:
Fog computing can be extremely convenient for offering power management as a service for
various networking systems. A service platform that can facilitate power management in
home-based IoT network with custom consumer control has been proposed in [34].
Moreover, it is also competent in managing the power usage of centralized Cloud data centres
in specific situations. The type of running applications determines the power consumed by
cloud datacentres. In this scenario, Cloud data centres can be complemented by providing
infrastructure for hosting several energy-extensive applications. Owing to the above, proper
power management will be ensured by reducing energy consumption in Cloud data-centres
[18]. Further, the emanation of carbon footprint can also be regulated by power management
[42].
The bottom sensing layer forms the basis of standard routing protocols. Failure to optimize
and handle communication resources is due to the shortage of control nodes. Contrastingly,
the self-organized coordinated algorithms are too complex for implementation.
Communication is frequently accomplished at the expense of node energy in data
transmission. Due to the demand for a large number of hops for transmissions, the reflection
to the dynamic network variation is consistently impeded. Despite that, the battery life is
inadequate and re-charging is out of the question owing to the small volume of sensor nodes.
The computation, storage and communication capabilities of the sensor nodes are all stifled.
Accordingly, along with frailer data aggregation performance, it takes loftier execution time,
with the likelihood of more errors [50].

Formations of clusters are important in communication in wireless sensors network. The data
should travel to optimal nodes and should be transferred through those paths using which
maximum efficiency can be achieved. This selection of optimal path is routing. Routing can
be performed using various software programmed devices known as protocols. Thus,
selection of optimal protocols is very crucial for network optimization and efficiency.
Routing algorithms have two types, that is non uniform and uniform clustering routing
algorithms. In a previous work [25], a non-uniform clustering routing algorithm was
proposed. The clusters, closer to the sink node, are smaller, and the number of sensor nodes is
less. The clusters away from the sink node, are larger and the number of sensor nodes is also
more. In this way, the energy consumption closer to the sink is greatly reduced. A method
based on fog computing was presented by the authors of[26]. These researchers stored some
data collected from the sensor nodes in the fog layer to avoid interruption in between
communication. This method also makes the movement of sink node easy. In another
work[27], an algorithm, which was used to calculate the neighborring node’s location, was
proposed. Filter based framework known as FERA was used. In this study, simulation was
used to observe the performance of the system. In [28]the amount of data that is transmitted
is decreased, which directly affects the lifetime of a network as the sensor nodes in WSN
consume maximum energy during transmission and reception of data packets. Author of [29]
presented a novel algorithm termed as complex alliance strategy with multi objective
optimization of coverage. A proportional relationship of the energy conversion function
between the working node and its neighboring nodes was also presented in this study. A self-
adaptive and self-organized protocol of routing algorithm LEACH was promoted by various
authors. Many of them worked on the betterment of protocol and some improved the LEACH
algorithm significantly [30]. Author of [31] exercise on forming cluster heads with balanced
energy by selecting two cluster head depending upon the energies and distance between head
and base station. Author of [32] experienced that optimistic distribution of computing load
shows better results as compared to the blind distribution. In this model clustering and
routing was done using Fuzzy Multiple Attribute Decision Making technique. A routing
scheme known as trust-based secure routing was proposed in a previous study [33]. The
source nodes send data and notification to sinks through disjoint paths, separately so that the
data and notification can be verified independently to ensure their security. This prevents
malicious attacks on the network and makes it more secure. Author of[24] proposed
clustering method based on mobile intelligent fog computing. This method is also known as
energy-efficient cross- layer-sensing clustering method (ECCM). The routing protocol is also
optimized using particle swarm optimization (PSO). All the previous proposed routing
algorithms have succeeded in transmitting the data effectively by choosing optimal routes but
all models also have limitations. High complexity and lack of energy efficiency are still one
of the major drawbacks of the available algorithms. The proposed algorithm uses fog
paradigms for optimizing the network. It introduces a fog layer for storing the data from the
sensors. The method is also optimized using various algorithms like Ant lion optimization
(ASO) and k - means clustering which enhances the efficiency of the network thus improving
the lifetime of the network.

In the era of IOT, importance of wireless sensor networks has increased. Energy consumption
in communication is dependent on the cluster formation. Selection of optimal routes and
nodes should be done in order to reduce energy consumption and latency. At present there are
many techniques available for cluster formation.

Over the years numerous algorithms were introduced to optimize cluster routes. Routing
algorithms can be classified into uniform and non-uniform clustering algorithms. Authors of
[25] proposed non uniform cluster routing algorithm. The objective of this algorithm was to
balance energy consumption in the nodes. In study [26] deployment scheme was proposed
which reduced long hops in cluster into multiple small hops to reduce transmission time in
the wireless sensor networks. Authors of [27] proposed a model based on fog layer
computing. This model stores partial information of the nodes in advance in the fog layer,
this reduces the interruptions between the communication and was helpful in moving sink
node. A low-cost collaborative charging in wireless sensor networks technique was presented
in study [28].

In [29] a data dissemination strategy which was based on ant colony optimization techniques
was proposed. Authors named this strategy as transmission with multiple load balancing
schemes (TMLBS). Load decentralization, load maintenance and load diversion schemes
were used in this proposed model. Researchers of [30] proposed a relative low-cost filter-
based framework FERA. This model adaptively used push/pull strategies and was able to
handle data request in VANET effectively. Authors of [31] proposed DMOA which was
motivated by morphological erosion and dilation on binary images. This algorithm was
capable of extracting event backbone nodes and estimates the event region on their basis. A
novel algorithm named as complex alliance strategy with multi-objective optimization of
coverage (CASMOC) which improves the congestion in network was proposed in [32].
Authors of this study also provided a proportional relation between working node and
neighboring nodes for energy conversion.

Due to the uncertainty of data, there could be coverage blind area and redundant data. To deal
with these authors of [33] proposed a novel model named as Coverage Control Algorithm for
Moving Target Nodes based on Sensing Probability (CMTN-SP). Survivability aware
connectivity restoration strategy was proposed in study[34]. This strategy involves the design
of load equilibrium mechanism and reliability enhancement measure. In study [35] a novel
algorithm called energy-based k-coverage control algorithm (EBKCCA) was proposed.
Based on the calculations by this algorithm authors concluded that multi transmission
consumes less energy than single transmission.

In study [36] Energy efficient method for clustering is proposed. Firstly, a strong node is
constructed in fog layer then data aggregation routing is constructed and at last particle
swarm optimization is used to elect cluster head for effective routing.

Above discussed algorithms have shown promising results but there is still some room for
advancement. Presently available algorithms are highly complex in nature. Apart from that
energy efficiency is always a reason of concern.
 Ideas Improvement Limitation
Article
Criteria
Nonlinear Power Suitable for centralized
Deng et
integers, Consumption, infrastructure, not easy to apply to
al.
Hungarian Delay fog computing
Method
Markov Power A continuous time queuing system
Guo et al.
decision consumption,
problems Delay
Heuristic User satisfaction, Complexity when number of user
Queis et
Task Latency, requests increases.
al.
Power
consumption
Table 2.2: Comparison of Power based optimization.
2.5 Security Concerns:
2.5.1 Authentication:
An essential feature of invasion defence in Fog based services is user authentication.
Unauthorised access in Fog services is unacceptable as they are employed in "pay as you go"
basis. Secured Fog computing environment requires user authentication, device
authentication, instance authentication, data migration authentication [25].
2.5.2 Privacy:
Data from end devices/sensors is processed in Fog computing. In certain situations, the data
closely associated with user activity and situation is detected. Hence, proper assurance of
privacy is considered a vital question in Fog computing. The privacy issues concerning Fog
based vehicular computing have been focussed upon for further examination, in [31].
2.5.3 Encryption:
Cloud computing is complemented by Fog computing. The data processed has to be
forwarded to Cloud in particular situations. It is indispensable to encrypt the data in Fog
nodes to protect sensitive information. A data encryption layer has been proposed in [23] to
take this into account.
2.5.4 DoS Attack:
It is challenging for Fog nodes to manage an extensive number of simultaneous requests as
they are resource constraint. Therefore, they can experience performance degradation.
Denial-of-Service (DoS) attacks play a critical role in creating severe service disruptions. Fog
nodes can be made busy for long durations of time by making a lot of unnecessary service
requests concurrently. Accordingly, resources for executing useful services become
inaccessible. This kind of DoS attack in Fog computing has been suitably examined in [24].
2.6 Research Gaps:
Following research gaps have been identified based on the literature review done in the
previous section:
1) To obtain an extra bit of battery life for fog devices has been one of the greatest
difficulties of fog computing solutions.
2) Researchers have used various optimization techniques for FOG Nodes but a very
less researchers have used Ensemble model approach followed by an optimization;
proposed work uses this as a Research Gap.
3) Aggregation of Fog nodes by migrating tasks from one node to another node can be
constructive in certain scenarios so as to manage power in a Fog environment.
4) The framework of Fog is distinct from Cloud so advancement of associated
programming languages to enable Cloud-based services and existing standards in
Fog are necessary.
5) Development of user interfaces and efficient networking protocols are necessary for
managing a large number of connections in Fog.
6) fog nodes needs improvement on availability, scalability, security, performance and
load balancing.
2.7 Problem Statement:

Industrial operators use the Industrial Internet of Things (IIoT) to automate the systems,
deploy sensors to monitor, measure, analyse data increase revenue opportunities and
improve efficiencies for manufacturing operations in wireless sensors network (WSN) has
let the manufacturing industry is experiencing substantial benefits. The quantity of data
from these plants can be computed in petabytes (1 million gigabytes) because of millions
of connected sensors on industrial control systems (ICSs), videos surveillance cameras
covering plants, millions of streaming, industrial robots, autonomous drones and so on.

Operational technology environments approached by traditional information technology

cannot meet the reliability security, network bandwidth difficulties, privacy and necessary
volume in a controlled operational environment. Hence, Fog environment is essential in
Wireless Sensors Networks (WSNs) to allow IIoT to reach its potential.

The approved data-based routing protocols for connecting Fog nodes are ordinarily
vulnerable to a huge quantity of energy and power voids in Wireless Sensors Networks. In
the problem formulation of this PhD research proposal, research gap # 3.1 is considered.
As per research gap 3.1, there is need to create Power-efficient framework for Fog
environment which can handle, migration of operators, dynamic load in the network and
migration of tasks.
2.8 Objectives:

The broad objectives of this PhD research work are:

1) To study and analyse existing Power Efficient resource utilization techniques in fog
environment.
2) To propose novel power efficient framework for resource utilization in fog
environment.
3) To implement and validate the proposed framework.
2.9 Methodology:

The objectives are fulfilled by using the following methodology:

1. The primary aim of this research work is to study the present available power efficient
algorithms in fog environment.
2. Based upon Fog resources like power consumption and QoS parameters a novel
resource utilization framework for Fog resources is proposed.
3. Framework is proposed after comparison with various other existing techniques.
4. Proposed framework is implemented using in MATLAB 9.4, 2020a.
5. Suitable architecture of sensor networks is used for the implementation in Fog
Environment.
6. The architecture of wireless sensor network is developed for Fog nodes.
7. The developed architecture is integrated with internet through specific protocols for
sharing of information and communications.
8. Further, simulations is performed to validate the suggested power efficient framework
for Fog environment and on the basis of post effects of proposed framework, a
sustainable solution is given.
2.10 Thesis Contributions:

1) A comprehensive survey of various Power Efficient frameworks is presented in this

work. This will help the researchers to be able to understand different Power Efficient
techniques.
2) The proposed model Power-Efficient Optimized Clustering Method (PEOCM) with
Intelligent Fog Computing for WSNs introduces a fog layer in WSN consisting of
cluster heads in the network. Sensing layer is optimized using ant lion optimization
(ALO), and the wireless sensors in fog layer are clustered using K- means clustering.
The proposed algorithm focuses on ALO only on the bottom layer and not on the fog
layer. To evaluate the performance of the proposed algorithm, the results are compared
with the findings of several traditional algorithms, such as energy-efficient cross-layer-
sensing clustering method, Distributed and Morphological Operation-based Data
Collection Algorithm (DMOA), and trust-based secure routing. Comparative results
showed that the proposed algorithm presented significant improvement.
3) The proposed model Machine Learning Based Power Efficient Optimized
Communication Ensemble Model with Intelligent Fog Computing for WSNs is based on
ant lion optimization which helps in initial cluster formation.
4) Further an ensemble model of k-means, SOM and fuzzy c-means is used to divide
clusters in FOG Nodes using weighted means algorithm.
5) Researchers have used various optimization techniques for FOG Nodes but a very less
researchers have used Ensemble model approach followed by an optimization;
proposed work uses this as a Research Gap.

Results shows significant difference in network life lime of sensor nodes when compared to
existing techniques.