1. EDGE computing and it’s role in architecting IoT
K. K. PATTANAIK
Wireless Sensor Networks laboratory
Atal Bihari Vajpayee –
Indian Institute of Information Technology and Management,
Gwalior
kkpatnaik@iiitm.ac.in
3. Academic programs offered
• 4 year BTech in CSE (started in 2017)
• 5 year Integrated Post Graduate in IT
• 5 year Integrated Post Graduate in MBA
• 2 year Mtech in CN, IS, DC, VLSI
• PhD
4. Figure: PDP loop explaining the ways in which Industry 4.0 might benefit
customers. Adapted from Center for Integrated Research, Deloitte US, 2018
• Shift from “linear data and
communications” to “real time
access to data and
intelligence” driven by the
continuous and cyclical flow
of information and actions
between the physical and
digital worlds
• Decentralization of digital
entities onto edge nodes
towards a modular structure of
MES
• Interoperability among digital
entities and execution of
sophisticated algorithms on
edge nodes.
Edge computing in Industry 4.0
5. Example on improving the availability of the
machine
Figure: OE without an adaptive mechanism Figure: OE as a consequence of adaptive mechanism
• Identification of “beginning of deterioration” and the “rate of deterioration” in OE of machine
is necessary to mitigate the unscheduled down state.
• Sophisticated algorithms at edge nodes to signal the deterioration and adaptively control the workload
will defer early breakdown and enhance average AOE
AOE =
𝐴𝑐𝑡𝑢𝑎𝑙 𝑡ℎ𝑟𝑜𝑢𝑔ℎ𝑝𝑢𝑡
𝐼𝑑𝑒𝑎𝑙 𝑡ℎ𝑟𝑜𝑢𝑔ℎ𝑝𝑢𝑡
Case of improving the average Absolute Operational Efficiency (AOE) of machine through edge intelligence
6. Visualizing Edge – An Introduction
Figure: Edge seen at different levels of granularity in an industry
• The Edge computing realm:
• Equipment protection
• Overall equipment effectiveness
monitoring
• Optimizing supply chain processes
• Predictive maintenance
• Improved performance: Alerts, analysis,
robustness, reliability, autonomous, resilient,
enhanced uptime
• Privacy and security: Data kept close to its
generation
• Reduced operational cost: Due to reduced data
migration, bandwidth, savings from cloud
spending
7. Visualizing Edge – An Introduction (continued)
• The in-device computing capability in real-time helps in
improving performances through embedded intelligence
in devices
• Huge IoT data generated is exploited at Edge
• Ratio of amount of data generated by all sensors to the
data used for decision is high
Edge Computing
Space
Cloud computing
Space
Low latency, increased privacy, less cost, real time processing,
relatively less processing capability, faster insights and actions,
improved response time, improved BW availability.
8. Visualizing Edge – An Introduction (continued)
Edge Computing
Space
Cloud computing
Space
Low latency, increased privacy,
less cost, real time processing,
relatively less processing capability,
faster insights and actions,
improved response time,
improved BW availability.
9. Abstract view of Edge paradigm
Source: IONOS Inc.
Source: IoT World Today
10. Figure: Layered view referenced to ISA 95 standard
Layered view of the ecosystem
11. Edge computing – Use cases
• Banks analyze ATM video feeds in real-time
• Mining companies use their data to optimize operations
• Chemical industries analyze the workers’ exposure trend to
harmful chemicals/gases
• Retailers can personalize the shopping experience of
customers and communicate specialized offers
• Industrial analytics
• Fleet management application
• Product traceability application
• Connected elevators to monitor the elevators’ health
• Cobots (collaborative robots) etc.
12. Data pro-sumers
• Data is both produced and consumed by us
• Traditionally all our data is placed on public/private cloud
and process them (term it as workload) on cloud
Cloud infrastructure
(Public/private)
Service providers
Network infrastructure
Service providers
Factory/workplace
(Data pro-summers)
Place workload as close to the place where the data is being produced
and action is being taken.
13. Data processing approaches
Figure: Classification of data processing mechanisms for IoT sensory environment.
Source: R K Verma et al., (2019), In-network context inference in IoT sensory environment for efficient network resource utilization, Journal of Network and Computer Applications,
Volume 130, 2019, Pages 89-103.
14. Non-message exchange based in-network data processing
Figure: Taxonomy of outlier detection techniques in wireless sensor networks.
Source: Bharti, S et al., (2016), Gravitational outlier detection for wireless sensor networks. Int. J. Commun. Syst. 29 (13), 2015–2027
15. Message exchange based in-network data processing
Figure: InContextIoT system architecture.
Source: R K Verma et al., (2019), In-network context inference in IoT sensory environment for efficient network resource utilization, Journal of Network and Computer
Applications, Volume 130, 2019, Pages 89-103.
16. Message exchange based in-network data processing
Figure: InContextIoT system architecture.
Source: R K Verma et al., (2019), In-network context inference in IoT sensory environment for efficient network resource utilization, Journal of Network and
Computer Applications, Volume 130, 2019, Pages 89-103.
Ex: high temperature, low humidity, high
luminosity, and presence of CO are the
LLCs acquired by annotating and tagging
the raw sensor data
LLCs are processed to infer HLC
Moving the process of the HLCs inference inside the
network and closer to the RoI of queries as against
the current approaches of centralized processing.
17. Edge server and edge devices
Edge compute capacity
(place where workload can be processed)
Edge server: A piece of IT
Equipment for processing IT
workload. It has relatively
higher processing abilities
Edge devices: A piece of IT
Equipment built for some purpose.
Ex: Assembling machine,
robot, car, quality check,
health of critical machine
Components etc.
19. How to manage workloads in these computing spaces?
• Reflection Questions:
How to manage a huge number of workloads in these diverse
compute spaces!
• Containerization
• Management of containers
• Security aspect etc.
Computing spaces
Edge devices’
Computing
space
Edge server
Computing space
Cloud computing
space
150+ billion edge devices by 2025
Computation will moved to edge
20. Enabling network protocols for edge computing
Low power, Low data, Short-range wireless mesh
network wireless standards
Z-Wave
ZigBee
Bluetooth LE
6LoWPAN
Industrial automation uses RS485 communication
protocol/PLC to monitor the device with Modbus
communication
Modbus
ONVIF
Open Network Video Interface Forum defines Interfaces
of physical IP-based security products for communication
21. Enabling network protocols for edge computing (continued)
SoC based self powered applications for the IoT
EnOcean
Open Platform Communications United Architecture is a
data exchange standard for industrial communication. It is
independent of the manufacturer of the application,
programming language, and the execution environment.
BACnet Building Automation and Control Networks used
to manage heating, ventilating, air-conditioning,
refrigerating, lighting, fire control, and alarm
systems.
OPC UA
The Long Range is a low power, networking protocol
designed to interoperate seamlessly between end devices
and the Internet in wireless manner
LoRa
22. Application layer protocols for gateway communication
To allow remote applications to communicate with the gateway
• MQTT: Message Queuing Telemetry Transport based on publish-subscribe
architecture
• AMQP: Advanced Message Queuing Protocol
• CoAP: Constrained Application Protocol, a web transfer protocol for IoT
• REST: Representational State Transfer architectural style for IoT atop
application layer
• WebSockets: A low-latency, full-duplex, persistent protocol that allows
the server to update the client application without an initiating request
from the client.
• JSON-RPC: A RPC protocol for microservices that allows clients to push
data/multiple calls to be sent to the server which need not be answered
in order.
23. What we are working on!
Figure: AgriCPS architecture
Source: Sapna et al., (2020), A dynamic distributed boundary node detection algorithm for management zone delineation in Precision Agriculture,
Journal of Network and Computer Applications, Volume 167
25. What we are working on! (continued)
• Edge server placement schemes to minimize edge device to server
communication cost
• Split learning in edge-cloud collaboration for predictive maintenance
• Optimizing communication cost for interactive IoT sensory environments
• Workload offloading in multi-access edge computing
Contributions of all the PhD scholars of Wireless Sensor Networks laboratory at ABV-
IIITM Gwalior is appreciated and duly acknowledged.
26. Connecting stuff: The IoT reference model
Source: http://cdn.iotwf.com/resources/72/IoT_Reference_Model_04_June_2014.pdf
Source: https://www.altexsoft.com/blog/iot-architecture-layers-components/