Edge Computing.pdf

Continuous Computing Plane
From Cloud to Edge
Remo Marconzini, 883256
Niccolò Puccinelli, 881395

Outline
Cloud Computing
Introduction to Edge Computing
Key concepts and architecture
Fog Computing
Orchestration and the case of Cloudify
Federated Learning
About Data Science
Future prospects

CLOUD COMPUTING
Cloud computing is a model for enabling
convenient, on-demand network access to
a shared pool of configurable computing
resources (e.g., networks, servers, storage,
applications, and services) that can be
rapidly provisioned and released with
minimal management effort or service
provider interaction (NIST, 2011).

NIST REFERENCE
This cloud model promotes availability and is
composed of:
High scalability and elasticity
High availability and reliability
High manageability and interoperability
High accessibility and portability
High performance and optimization
5 essential characteristics:
Software as-a-Service (SaaS)
Platform as-a-Service (PaaS)
Infrastructure as-a-Service (IaaS)
3 Service Models:
Public Cloud
Private Cloud
Hybrid Cloud
Community Cloud
4 Deployment Models:

Cloud is designed to be: highly realiable, highly scalable and it
provides elasticity for computing, storage and network resources.
Cloud provides business benefits, such as an on-demand pay-as-
you-go service to the users, which lowers the owning cost for the
companies (from Capex to Opex).
Cloud architectures address the difficulties related to the large-
scale data processing.
"GOOD AND BAD"
WITH CURRENT CLOUD COMPUTING MODEL
We can simply deliver all the computing services such as networking,
database and analytics over the internet...

... however, such a centralized model will face significant challenges for the IoT world.
The Internet is evolving rapidly towards the future Internet of Things (IoT), which will potentially
network billions or even trillions of devices.
Due to this, the current cloud computing model will have to deal with:
"GOOD AND BAD"
WITH CURRENT CLOUD COMPUTING MODEL
Volume and Velocity of data accumulation
of IoT devices.
Security and Privacy concerns.
Latency due to the distance between edge
IoT devices and Datacenters.
Scalability challenges.
The costs of storing, processing and analysing
data from IoT devices.

WHY IS THE CURRENT MODEL NOT ENOUGH?
Applications that require Autonomy.
Industry 4.0.
Self-driving car.
Completing tasks with little or no human interaction.
Applications that can’t tolerate Latency.
Health Care.
Financial transactions.
Latency can be reason for failure.
Applications that require significant Bandwidth.
Smart surveillance system.
Traffic management.
Generate significant amount of data.

AND EDGE COMPUTING CAN...
Reduce response times
by processing data at
source.
Reduce data transport
costs to the cloud
center.
Improve the
reliability of many
operations by being
able to work without
a connection.
Improve security by
keeping data at
source.

WHAT IS EDGE COMPUTING ?
Edge computing is a distributed IT architecture that
moves the compute power physically closer to where
data is generated, usually an IoT device or sensor.
Named for the way compute power is brought to the
edge of the network or device, edge computing allows
for faster data processing, increased bandwidth and
ensured data sovereignty, in contrast to Cloud
computing where the computation takes place in a
central location, increasing the associated costs and
response time.

WE NEED MORE SPEED...
Businesses use edge computing to improve the response time of
their remote devices. Edge computing makes real-time computing
possible in locations where it would not normally be feasible.
In order for a car to drive down any road safely, it
must observe the road in real-time and stop if a
person walks in front of the car. In such a case,
processing visual information and decision
making are performed at the edge.
Centralized cloud systems introduce delays in data transmission
caused by network latency...

WE CAN'T TOLERATE LATENCY...
The healthcare industry is drowning in data. The low latency, mobility
and data processing capabilities of edge computing at a network
edge can enable faster and more accurate diagnosis and treatment.
Edge computing can improve emergency services by
allowing paramedics to make faster and more
accurate diagnoses on-site, and providing
hospitals with more detailed information on
incoming patients.
In a centralized cloud system, the execution of these applications is often
perceived as insecure due to the sensitivity of patient data...

Cloud computing. Would have to send high definition
video/images to the cloud to know how many parking
spaces are available. Higher bandwidth.
Edge computing. Should only send to the central cloud
the number of available parking spaces. Bandwidth
reduction.
In a Smart Park System scenario...
TOO MUCH BANDWIDTH!
Computing at the Edge means less data transmission as most of the heavy-
lifting is done by the edge devices. Instead of sending the raw data to the
Cloud, most of the processing is done at the edge, and only the result is sent
to the Cloud.

MILESTONES OF EDGE COMPUTING
Akamai, Content delivery
network
Nobel et al, Agile application-
aware adaptation for
mobility
Amazon, Elastic Compute
Cloud
Satyanarayanan et al., The
case for VM-based cloudlets
in mobile computing
Cisco, Introduced the term
"fog computing"
Proximity-routing in
overlay P2P networks

WHAT DOES MARKET SAY ?
https://www.grandviewresearch.com/industry-analysis/edge-computing-market
CAGR of 33.4% from 2022 to 2030.
In terms of components, hardware
holds 40% of the market in 2021.
In terms of applications, IoT holds 28%
of the market in 2021.
According to Gartner, “Enterprises that have deployed edge use cases in production
will grow from about 5 percent in 2019 to about 40 percent in 2024.”
https://www.gartner.com/en/documents/3986290

KEY CONCEPTS OF EDGE COMPUTING
Real-time data processing with low latency.
Geographically distributed.
Reliable & resilient.
Data sovereignty.
User interaction.
IoT.

REAL-TIME DATA PROCESSING
Latency is the time needed to send data between two points on a network. Although
communication ideally takes place at the speed of light, large physical distances
coupled with network congestion or outages can delay data movement across the
network.
By distributing servers and storage where the data is generated, edge computing
can handle many devices using bandwidth exclusively from the local devices
generating the data, making latency virtually non-existent.
Local servers can perform essential edge analysis and data pre-processing to
make real-time decisions, before sending the essential data to the central cloud.

RELIABLE & RESILIENT
Edge computing continues to operate even when communication channels are
intermittently available or temporarily down (i.e. edge computing deployments on
an oil rig).
Edge computing doesn't necessarily require a constant connection to transmit data to
the central cloud; instead, it can choose to move only the processed information to
the central cloud when the connection is available.
Edge computing also improves resilience by reducing the central point of failure,
improving the reliability of the entire connected environment.

DATA SOVEREIGNTY
Moving huge amounts of data isn't just a technical problem. Data's journey across
national and regional boundaries can pose additional problems for data security,
privacy and other legal issues.
Edge computing can be used to keep data close to its source and within the bounds
of prevailing data sovereignty laws, such as the European Union's GDPR.
This can allow raw data to be processed locally, obscuring or securing any sensitive
data before sending anything to the cloud or primary data center, which can be in
other jurisdictions.

For example, an IoT device such as a smart thermostat can
use edge computing to process sensor data locally and make
adjustments to the temperature in near real-time, without
needing to send that data to a central server for processing.
In terms of interacting with users, edge computing can enable
devices such as smartphones and IoT devices to perform
processing tasks without needing to constantly send data back
and forth to a central Cloud. This can lead to a better user
experience, as well as improved security and privacy.
USER INTERACTION & IoT

EDGE COMPUTING ARCHITECTURE
Edge computing hardware often consists of several physical components...

EDGE COMPUTING HARDWARE & NETWORKING
The edge and IoT devices are equipped to run
analytics, apply ML algorithm, and even store
some data locally to support operations at the
edge.
They typically have limited computational
resources
Edge devices

Routers are edge devices that connect networks.
For example, a router at the edge may be used to
connect an enterprise's LANs with a WAN or the
internet.
Switches, which are also referred to as access
nodes, connect several devices in order to create
a network.
IoT devices, like smart cameras, thermometers,
robots, drones, vibration sensors ecc...
Edge devices

Gateways are a crucial part of edge computing
architecture that extends cloud capabilities to
local edge devices and performs essential network
functions like:
enabling wireless connectivity.
providing firewall protection.
processing and transmitting edge devices
data.
Gateway

Gateway
It's like having a mini cloud with efficiency, security, low
latency and local autonomy.
They are in constant communication with the devices by
using agents that are installed on each of the devices.
Million of gateway maintain a pulse on the billions of
devices.
Some edge devices can serve as a limited gateway or
host network functions, but edge gateways are more
often separate from edge devices.

Edge Cloud
New networking technologies have resulted in the
edge network or edge cloud, which can be
viewed as a local cloud for devices to
communicate with.
The edge Cloud reduces the distance that data
must travel from the devices and thus decreases
latency and bandwidth issues.

Some HW characteristics
Temperature resistant: Edge hardware is often placed
outside in freezing, sweltering, and wet climates.
Tamper resistant. Because edge computing devices
are often used in far-flung locations where they cannot
be consistently monitored.
Protected from cyberattacks: Edge devices tend to
be more vulnerable to bad actors. So, they must be
equipped with security tools like firewalls and
network-based intrusion detection systems.

CONTINUOUS COMPUTING PLANE
New architectural approach that allows for data and workloads to be processed and stored in
various locations ranging from the cloud to the edge.
Multiple benefits
coming from
different
environments

How to integrate Cloud and Edge computing?
Combining strengths of both technologies for a more efficient and effective computing system.
Common issues:
High-latency communication between centralized cloud and edge devices must be
minimized.
Slow resource management and scaling.
Edge devices low-computational resources (e.g. sensors, smartband) can't store and
process complex real-time data.
Bottleneck between edge devices and cloud: if all (or most) devices are connected to the
same cloud data center, the network needs a very high bandwidth to avoid the bottleneck.
Security and privacy issues in data transfer.
Need for an intermediate layer.

FOG COMPUTING
Intermediate layer between cloud and edge computing.
Edge clouds (nodes)
Distributed infrastructure that places cloud
computing resources closer to the edge of the
network (sort of small data-centers).
Provide low-latency access to data and services for
devices and users that are located near the edge of
the network.
Equipped with processing, storage and networking
capabilities.
Data filter between edge and cloud.

Scalability
Distributed architecture: additional resources can
be added closer to where they are needed.
Decentralized management: resources managed
and controlled at the edge, so they can be added or
removed without affecting the entire system.
Dynamic resource allocation based on the needs of
users and devices.
Offloading to the cloud (hybrid architecture).

Cloud offloading
Transferring certain computing tasks or data to a
more powerful resource for processing.
To nearby fog nodes:
Tasks requiring low latency (e.g. real-time video
processing)
Data that are too large to be transmitted over
long distances.
To a central cloud:
Tasks that require more powerful processing
capabilities, such as machine learning or big data
analysis.

Common algorithms
Static: predefined rules based on the features of
the application and the device resources.
Dynamic: based on the current state of the
device resources, the network conditions and the
processing requirements of a specific task.
Hybrid: combines static and dynamic strategies.
ML-based: uses machine learning algorithms to
analyze data and make offloading decisions.
Improved performances and scalability, as it allows
for more efficient use of resources.

Delay tolerant
Ability of the system to continue functioning even in
presence of network delays or disruptions.
Achieved through:
Local processing.
Redundancy: multiple fog nodes and devices.
Offloading: transfers certain computing tasks or
data in case of network failure or delay.
Scheduling and prioritization.

Cost effectiveness and efficient energy consumption.
Data processed closer to the source.
No need for continuous transmission to centralized
data-center.
Reduced dependence from centralized cloud
resources.
Greener computing.

Mobility
Ability of the system to adapt to changes in the
location and movement of devices and users.
Achieved through:
Handover management: continuity of service
and low-latency connections as devices move
between different fog nodes.
Context awareness: providing services tailored
to the specific context of the user with the
support of IoT devices.
Mobile edge computing (MEC): MEC servers
can provide low-latency, high-bandwidth access
to data and services for mobile devices
(especially with 5G).

Need for an orchestration system.
Distribution and coordination of resources, workloads and applications across different
computing environments.
Metrics defined by SLA: bandwidth, users
Different approaches.
Centralized.
Local.
Distributed.
needs, processing capabilities, ...
ORCHESTRATION

Cloud orchestration domain
Cloud characterized by:
Centralization. Many hosts in few central locations.
Virtually unlimited pool of resources. Compute, network, and storage resources viewed
as unlimited and capable to serve multiple clients at one time.
Fast network connectivity within data-centers.
So we need:
Rapid elasticity. Steady and predictable performance at the lowest possible cost.
Moveable workloads to solve healing and scaling issues.
Data-center security.

Edge orchestration domain
Edge site characterized by:
Distribution. Few hosts in many distributed locations.
Limited resources.
Network limitations. Frequent network outages.
So we need:
Autonomy. Offline operability for a long period of time.
Resilient networking. Services designed for network outages and slow connections.
Static workloads. Workloads have to stay where they are deployed, unless offloading.
Local security.

Centralized orchestration
Single centralized controller to manage and coordinate the deployment and operation of edge
devices and services.
Pros:
Easier monitoring.
Security: access control and policies.
Cons:
Single point of failure (also security issue).
Limited scalability, not suitable for highly
distributed systems.
Latency and bandwidth issues as the
number of devices and services increase
(bottleneck).

Local edge orchestration
Edge devices and services to manage and coordinate their own deployment and operation, with
minimal centralized control.
Pros:
Decentralization and locality.
No single point of failure.
Reduced latency.
Autonomy, but devices need to be equipped with the
necessary software and algorithms.
Cons:
Complex integration and interoperability.
Limited visibility and control: it's difficult to monitor and
control the behavior of the system.

Distributed orchestration
Distributed network of controllers to manage and coordinate the deployment and operation of
edge devices and services.
Benefits from both central and local orchestration.
Pros:
Scalability and flexibility, suitable for highly
distributed systems.
Reliability and redundancy provided by multiple
controllers.
Low latency for local real-time operations.
Security: no single point of failure and policies control.
Cons:
Complex synchronization and interoperability.

Open-source unified platform for management, deployment and automation of applications
and services across cloud and edge environments.
3 main components communicating with each other:
Management console.
Orchestration engine.
Light-weight local agents and plugins.
The case of Cloudify

Management console
User web-based interface.
Managing the entire application life-cycle, including provisioning, scaling and monitoring.
Provides status and performance metrics of the edge devices.
Communicates with the other components.

Orchestration engine
Automating deployment and management of applications and services on the edge devices.
2 ways to specify the desired configuration of edge devices.
Pre-defined templates with pre-defined policies for scaling and updates.
Custom blueprints.
Cloudify blueprints.
YAML files defining components (such as VMs,
load balancers and DBs), relationships and
policies that make up an application or a service.

Local edge agent
Autonomous local edge lightweight orchestrator.
Can support a virtualization layer or run on a bare-metal machine.
Can work with a container orchestration system like Kubernetes or orchestrate the box directly.
Communicates with management console and orchestration engine.
Plugins to extend the functionality of the platform.
Infrastructure plugins to integrate Cloudify with
infrastructure providers for virtual resources (e.g.
AWS, Microsoft Azure).
Application plugins to integrate Cloudify with
software systems and tools, such as databases or
messaging systems.

FEDERATED LEARNING
Distributed ML approach developed by Google in 2016.
Exploiting edge computing

Data storage and model training on server cloud.
Data must be transmitted from the edge to a central data-center, in order to train the model.
Centralized approach
Drawbacks:
Communication costs. We need a very high
bandwidth to avoid bottleneck.
Reliability. Performance degradation and
failures due to wireless connection.
Security and privacy. Data could be used
inappropriately by companies or could be
stolen in an attack to the centralized
infrastructure.

Co-training between different devices while maintaining locality.
Edge devices do not share their local dataset, only the model training parameters.
Distributed approach
Advantages:
We use local data, which are not frequently sent to
a remote server.
Reduced latency.
Reduced bandwidth required.
Improved reliability.
Improved privacy and security.
Very suitable for edge computing.

The central/edge server determines an ML model to be trained locally.
A subset of clients is chosen (randomly or through algorithms such as active learning, wherein
clients are selected based on the uncertainty of their predictions).
The server multicasts the model to selected clients, who download the global parameters and
train the model locally.
Each client sends its own local model updates to the server.
The server receives the updates and aggregates them using aggregation algorithms (e.g.
FedAvg, which averages model updates across all clients) to generate a new global model.
1.
2.
3.
4.
5.
FL steps
The server never accessed client data!

WHAT ABOUT DATA SCIENCE
Image classification.
Self-driving cars.
Surveillance cameras.
Object detection.
Tracking people or vehicles.
Natural Language Processing.
Improving text-to-speech application.
Improving voice commands processing.
REAL-TIME!

THE FUTURE OF EDGE COMPUTING
Increased use of edge computing in IoT applications, as more and more devices
become connected to the internet.
Real-time data analysis and decision making in several areas (e.g. healthcare).
Advancements in edge hardware.
More powerful edge devices and larger batteries.
Development of new edge computing platforms.
Easier implementation of edge computing solutions for businesses and organizations.
Improved data security, as edge computing and ML become more popular.
Better security features and ML algorithms to detect and prevent security breaches.

REFERENCES
X. Wang, L. T. Yang, X. Xie, J. Jin and M. J. Deen, "A Cloud-Edge Computing Framework for Cyber-
Physical-Social Services" in IEEE Communications Magazine, vol. 55, no. 11, pp. 80-85 2017.
J. Ren, Y. Pan, A. Goscinski and R. A. Beyah, "Edge Computing for the Internet of Things" in IEEE
Network, vol. 32, no. 1, pp. 6-7, 2018.
J. Pan and J. McElhannon, "Future Edge Cloud and Edge Computing for Internet of Things
Applications" in IEEE Internet of Things Journal, vol. 5, no. 1, pp. 439-449, 2018.
Shabnam Kumari, Surender Singh, Radha , " Fog Computing: Characteristics and Challenges" ,
International Journal of Emerging Trends & Technology in Computer Science (IJETTCS) , Volume 6,
Issue 2, March - April 2017 , pp. 113-117.
Xu, Z., Zhang, Y., Li, H. et al. Dynamic resource provisioning for cyber-physical systems in cloud-
fog-edge computing. J Cloud Comp 9, 32 (2020).
Dante D. Sánchez-Gallegos, Diana Carrizales-Espinoza, Hugo G. Reyes-Anastacio, J.L. Gonzalez-
Compean, Jesus Carretero, Miguel Morales-Sandoval, Alejandro Galaviz-Mosqueda, From the edge
to the cloud: A continuous delivery and preparation model for processing big IoT data,
Simulation Modelling Practice and Theory, vol. 105, 2020, pp. 102-136.
Sulieman, N.A.; Ricciardi Celsi, L.; Li, W.; Zomaya, A.; Villari, M. Edge-Oriented Computing: A Survey
on Research and Use Cases. Energies 2022, 15, 452.
Abreha, H.G.; Hayajneh, M.; Serhani, M.A. Federated Learning in Edge Computing: A Systematic
Survey. Sensors 2022, 22, 450.

REFERENCES
https://blogs.nvidia.com/blog/2022/01/05/difference-between-cloud-and-edge-computing/
https://towardsdatascience.com/you-need-to-move-from-cloud-computing-to-edge-computing-
now-e8759eb9690f
https://azure.microsoft.com/en-us/resources/cloud-computing-dictionary/what-is-edge-
computing/#:~:text=With%20edge%20computing%2C%20businesses%20can,the%20location%20
where%20it's%20collected
https://www.slideshare.net/chetansk/edge-computing-109820603
https://www.accenture.com/us-en/insights/cloud/edge-computing-index
https://www.stackpath.com/edge-academy/what-is-edge-orchestration/
https://cloudify.co/blog/birth-of-edge-orchestrator-cloudify/
https://techleadcuriosity.com/edge-computing-and-machine-learning-unlocking-the-potential-
of-real-time-data-analysis-b53876530d15
https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication500-292.pdf

REFERENCES
https://slideplayer.com/slide/13778757/
https://www.volico.com/whats-driving-the-adoption-of-edge-computing-in-the-healthcare-sector/
https://blog.bosch-si.com/bosch-iot-suite/cloud-and-edge-computing-for-iot-a-short-
history/#:~:text=The origin of edge computing,such as images and videos.
https://www.grandviewresearch.com/industry-analysis/edge-computing-market
https://www.techtarget.com/searchdatacenter/definition/edge-computing
https://www.techtarget.com/searchcloudcomputing/tip/Explore-edge-computing— services-in-
the-cloud
https://www.techtarget.com/searchnetworking/answer/What-is-the-network-edge-and-how-is-it-
different-from-edge-computing
https://developer.ibm.com/articles/edge-computing-architecture-and-use-cases/
https://www.ibm.com/cloud/architecture/architectures/edge-computing/

Edge Computing.pdf

More Related Content

What's hot (20)

Similar to Edge Computing.pdf (20)

Recently uploaded (20)

Edge Computing.pdf