FOG COMPUTING- Fog

Shifting from Cloud to

Abstract— To support the computational demand of devices interact with one another and support the data
real-time latency-sensitive applications of largely geo- communication within the Internet of Things.
distributed IoT devices/sensors, a new computing
paradigm named Fog computing has been introduced. The term IoT is a giant network of connected
Genera lly, Fog computing resides closer to the IoT "things" (which also includes people). The
devices/sensors and extends the Cloud-based computing, relationship will be between people-people, people-
storage and networking facilities. These devices, called things, and things-things. Fog computing is a
fog nodes, can be deployed anywhere with a network distributed computing paradigm that acts as an
connection: on a factory floor, on top of a power pole, intermediate layer in between Cloud data centers and
alongside a railway track, in a vehicle. Any device with
IoT devices /sensors. Here the term edge refers to
computing, storage, and network connectivity can be a
fog node. Examples include industrial controllers, different nodes to which the end user is connected
switches, routers, embedded servers, and video and it is also called edge computing. So we can
surveillance cameras. conclude that Fog=Internet of Things + Cloud. It offers
to compute, networking and storage facilities so that
Keywords— Cloud computing, computation
offloading, edge computing, fog computing, internet of Cloud-based services can be extended closer to
things, mobile cloud computing, mobile edge computing the IoT devices/sensors. [1] Since Cloud data centers
are geographically centralized, they often fail to deal
with storage and processing demands of billions of
INTRODUCTION geo-distributed IoT devices/sensors. Typically, a Fog
Now a day, most information is accessible on computing environment is composed of traditional
mobile devices. Social media data, emails, files and networking components e.g. routers, switches, set top
even bank account information is easily available on boxes, proxy servers, Base Stations (BS), etc. and can
mobile devices simply with an internet connection. Yes, be placed at the closer proximity of IoT
the Cloud is making it all happen. However, slow devices/sensors. Consequently, the networking
bandwidth is covertly contributing to the increase in components enable Fog computing to create large
restrictions of a wireless network. 3G and 4G cellular geographical distributions of Cloud-based services.
networks aren’t efficient enough to transmit data from Besides, Fog computing facilitates location
devices to the Cloud at the same speed that the data is awareness, mobility support, real-time interactions,
generated. Thus, Cisco decided to store and process scalability and interoperability. Mobility is
this volcano of data on the routers that are present compromised due to the huge number of nodes. The

between the Internet and end user devices and deliver main task of fog is to deliver data and place it closer
content to the end users through a more geographically to the user who is positioned at a location which at
distributed platform. This new methodology of edge the edge of the network.
computing is focused on distributing data by moving it
closer to the end user to remove latency and sustain
mobile computing. [3] The term fog computing is
given by CISCO as a new technology in which mobile
 The remaining sections of this analysis are as follows: open nature makes it vulnerable to a broad variety of
The second section is a literature review, and it contains a hazards. The open nature of the industrial setting makes it
synopsis of the greatest new research that were looked at. hard to spot new hazards. It causes defences that prevent
Part 3 lays out the model and architecture of the system, malicious outsiders from wreaking devastation on human
describing in detail how it is put together. In Section 4, we systems to fail. That's why it's so important to spot signs of
provide our extensive experimental findings, which include a hostility. When it comes to providing an efficient model, ICS
performance trajectory and a comparison to existing models. offers a revolutionary hybrid deep learning approach. Initial
The work presented here concludes in Section 5. data gathering will focus on the ICS industry. The second
step, pre-processing, involves eliminating noise and other
I. RELATED WORKS anomalies from the raw data. Finally, the Hybrid Honey
Golchha et al. [15] state that a multi-layered strategy is Badger-World Cup Algorithm is implemented to choose
required to ensure the care of the industrial system. An (IDS) optimal features from the raw data itself (HHB-WCA). In the
may examine the data packets in an IIoT setting to spot any hybrid deep learning model known as Auto-encoder-Bi-
attempts at hacking. In this research, we present an IIoT directional encoded features from Auto-encoder are retrieved
cyberattack detection method framework, we combine state- and presented to the Bi-LSTM classifier, with the outputs
of-the-art CatBoost with more traditional ones like classifier being the selected optimal features. Once the simulation
to effectively identify cyberattacks. CatBoost has the highest results are in, the suggested method is compared to the other
accuracy (99.85%) of the models we recommend, followed old-style approaches in order to find the hidden lessons in the
by HGB (97.90%) and RF (98%3%). network.
Sharma et al. [16] offer an (OFL-SRPL) that can run on Alazab et al. [19] established a wide variety of techniques
AMI hardware. The OFL-SRPL employed the (PSO) for detecting routing assaults. These techniques leverage
technique to choose the best local perfect from the HAN different facets of network data. This study examines the
model while minimising the quantity of data sent between prevalence of routing attacks against the Power and Lossy
the multiple local models and the server. The proposed OFL- Networks (RPL). When it comes to IoT systems based on
SRPL uses a collection of sequential classifiers, each with its 6LoWPAN, RPL is essential. In this research, we outline
own optimal loss function, to improve decision quality while threats that either originate in, or are adapted from, Wireless
reducing the burden on FL of sending ensemble model using Sensor Networks by exploiting RPL-specific vulnerabilities
consecutive classifications and only transmits the weight of (known as RPL-specific attacks). The article describes a
the best model to generate the global model. Using a wide variety of cyberattacks, including flood attacks, data
sequential classifier layout, we are able to address class rank assaults, blackhole attacks. In order to detect routing
imbalance issues and enhance the ownership weight. After attacks, this study offers a decision tree classifier with a
achieving the best global model selection possible, devices Support Vector Machine classifier that only uses one class.
efficiently execute ensemble classifiers with optimal learning The HIDS combines the best elements of a System (IDS) in
limits. As a result, using the proposed OFL-SRPL improves order to identify routing attacks with high accuracy and a
detection efficiency by 10.6 percentage points. low false alarm rate (AIDS). The proposed HIDS was tested
using real-world IoT network traffic and several routing
Traditional FL can achieve similar results with fewer attacks taken from a routing dataset to gauge its efficacy.
iterations of communication than a centralized LSTM model Both are improved with the proposed hybrid IDS compared
arrangement. to those of SIDS and AIDS techniques, respectively.
Rashid et al. [17] propose a for detecting unauthorized Abosata et al. [20] offer a concept for a customized
intrusions into IoT networks. Federated training utilizing distributed intrusion detection system (FT-CID) to detect
data from neighbouring IoT devices ensures both privacy and RPL infiltration in a varied IoT. In order to create an
security. A centralized global server disseminates a superior effective FT-CID, it is necessary to collect data, train edge
detection mechanism, with local IoT clients' sole input being IDS with the aid of FTL, and detect intrusions. A local model
modifications to the server's parameters. After each training, is trained using data acquired from RPL-IoTs after a central
the global server sends to each IoT client, which then trains server loads a specified learning model into the FT-CID.
its own dataset in order to preserve the privacy of individual Second, is produced via and the edge provided by the central
IoT devices while simultaneously optimizing the model. To server. Thus, we employ transfer learning to refine the server
evaluate the efficacy of the projected method, we fabricated and edge settings with relationship information. In order to
a new dataset (named Edge-IIoTset) and subjected it to a build and fine-tune an IDS model using partial retraining,
battery of tests. The performance evaluation shows that the local learning is based on the information saved on servers
suggested intrusion detection model is reliable and effective, all over the world. Finally, the FT-CID model's unique IDS
with an accuracy of 92.49% utilizing the FL technique, detects and reports on intrusions across a wide range of IoT
which is on par with that of conservative centralized ML networks. Also, the FT-CID architecture uses FTL to provide
models (93.92%). high RPL security by tacitly leveraging the local and global
Signature-based identification uses anomaly detection, as features of a number of IoTs. The FT-CID demonstrated an
explained by Nagarajan et al. [18]. Despite this, the ICS is accuracy of 85.52% in detecting RPL intrusions when tested
constructed using a large number of different software on a heterogeneous IoT network.
applications and hardware parts. It would appear that the
II. PROPOSED SYSTEM
The suggested model for an ensemble of IoT devices is
an anomaly-based intrusion detection scheme that can
automatically create profiles for benign and harmful