Making Actionable Decisions at the Network's Edge

Making Actionable
Decisions at the
Network’s Edge
In the evolving hyper-connected world of the Internet of Things,
immense new possibilities are emerging from interlinked
ecosystems that can make fast, actionable decisions uncon-
strained by traditional analytical processes. Edge analytics is
fast emerging as a way of extending the limits of cloud-enabled
decision-making.
November 2017
DIGITAL SYSTEMS & TECHNOLOGY

22
Digital Systems & Technology
| Making Actionable Decisions at the Network’s Edge
EXECUTIVE SUMMARY
The Internet of Things (IoT) is enabling innovative new connected ecosystems to emerge
that amplify the value organizations provide to end customers. Smart cities, smart infra-
structure, driverless cars and real-time guidance using location intelligence are just a few
of the ways these new solutions touch our everyday lives. However, what makes such con-
nected ecosystems tick, succeed and evolve is their ability to process in near-real time
the huge amounts of data generated at the network’s edge by sensors and instrumented
devices.
The emergence of the cloud has made building such ecosystems possible, but addition-
al time and enhanced connectivity is still required to inform fact-based decisions after
analysis. Sending data to the cloud and awaiting analytical results costs precious millisec-
onds, damaging real-time responsiveness. Billions of devices produce data persistently, but
managing and making sense of this big data requires a huge investment in computational
analytics, storage and networking software, as well as powerful computing platforms. Or-
ganizations that hue to traditional approaches will be unable to tap the IoT’s full potential.
A new and evolving technique — commonly known as edge analytics — is rapidly emerging
as the go-to mechanism for overcoming existing infrastructure limitations. While edge ana-
lytics derives from cloud analytics, it goes one step further by democratizing the ability to
analyze data — not just in cloud data centers but at the point of data collection, on devices
themselves and in the gateways that interconnect enterprise ecosystems.
Edge analytics has reached a major inflection point. In data-warehouse-oriented analytics,
data is typically sent, via gateways, to cloud systems where the entire analytics process

33Making Actionable Decisions at the Network’s Edge |
occurs. Data visualization is presented after a series of steps performed in the cloud. While
this works fine in situations that do not require decisions to be made quickly, digital-era
demands are quickly changing the business calculus.
In edge analytics, data is collected and analyzed close to the source of data generation.
So, instead of sending data to the cloud for analysis and then waiting for a response, edge
analytics brings more computation to the edge, saving on time as well as cost of data trans-
mission. Simple model-based analytics can be conducted on the device/sensor itself while
more complex analytics that require data from multiple devices can be performed on IoT
gateways, and finally the most sophisticated form of analytics — commonly called big data
analytics — can be handled on the cloud. This analytics hierarchy reduces the complexity
and burden on the network and the data centers. Distributing the analysis of data to the
edge is a powerful way of unlocking IoT value.
Edge analytics is not just about gaining operational efficiencies or making the business
more scalable. Many businesses do not require complex or sophisticated data analytics, but
do need speed and automation. Any delay in delivering results or loss of data due to con-
nectivity failure can cause reputational and even financial damage to the organization. It is
another reason why gateways need to be placed closer to the source so the data generated
can be cleaned, batched and sent back to decision-makers in shorter timeframes.
This white paper explores how to build a connected ecosystem that not only has a brain in
the cloud, but also reflex actions at the edge. It also offers some case illustrations to help
decision-makers envision edge analytics’ art of the possible.

4
THE CLOUD ON EDGE
Large organizations have benefitted from the concept of cloud computing. But as the IoT grows, along
with the amount of the data generated to inform real-time decision-making, it is essential to access
this data quickly — but without incurring huge investments of time and money. This is where edge ana-
lytics comes into play. The trick is to incorporate both models to their best effectiveness: deploy edge
analytics where time is of the essence, and use cloud analytics where security and data volumes are
the deciding factors. It is imperative that IoT strategies make use of the best of both cloud computing
techniques and edge analytics processing to optimize IoT ecosystems (see Figure 1).
Edge vs. Cloud
Cloud Analytics Edge Analytics
Cloud
Internet
Network
Devices
Data
Data
Cloud
Internet
Network
Edge
Devices
Data
Data
Figure 1

Making Actionable Decisions at the Network’s Edge | 5
The trick is to incorporate both
models to their best effectiveness:
deploy edge analytics where time
is of the essence, and use cloud
analytics where security and
data volumes are the deciding
factors. It is imperative that IoT
strategies make use of the best of
both cloud computing techniques
and edge analytics processing to
optimize IoT ecosystems.

6 | Making Actionable Decisions at the Network’s Edge
In the IoT age, nearly every connected and
instrumented devices generates huge amounts
of data. The underlying metadata, however, is
useless unless it is analyzed for meaning.
AN INTRODUCTION TO EDGE ANALYTICS ARCHITECTURE
The hierarchy of edge analytics can be represented as a three-tiered architecture (see Figure 2, next
page). The flow of data begins with sourcing of raw data from smart devices or sensors followed by
more sophisticated analysis on gateways at the edge of the network and finally some “heavy lifting,”
or big data analysis, using complex cloud computing models.
Tier 1: The Sourcing of Raw Data
In the IoT age, nearly every connected and instrumented devices generates huge amounts of data.
The underlying metadata, however, is useless unless it is analyzed for meaning. Much of the data
collected does not require complex analytics, hence data from these devices can be analyzed on the
“edge” - i.e., close to the source of data generation – to deliver near-instant automated results.
Tier 2: Processing Data on the Network’s Edge
Edge analytics deploys gateways on the edge of the network. These gateways connect, collect and
analyze data in near-real time. The outcome of this analysis can be transferred back to the devices
immediately or can be stored in a small, low-cost memory device. The stored data can be further
transferred or routed to the cloud for advanced analytics.
Distributing analytics on the network to different edge nodes has many advantages. It decreases the
complexity that companies face while computing huge amounts of real-time data and increases the
scalability by distributing the computation workload across multiple edge nodes.
Tier 3: Sophisticated Cloud Computing
Filtered data from the edge of the network is transferred to the cloud for more complex processing.
Data is sent to the cloud from multiple gateways to store, process or analyze. Generally, data that does
not require an instant response is transferred to the cloud for heavy-duty processing.
EDGE ANALYTICS BUSINESS DRIVERS
With the deluge of data has come the need to use it quickly and in real time. IoT devices pump out
data in huge amounts and with the increasing number of smart connected devices, the need for edge
computing is clear. According to Gartner, there will be 20.4 billion IoT devices by the year 2020.1
In this
scenario, it is important to identify which business drivers spur a demand for ongoing edge analytics.

Three-Tier Edge Analytics Architecture
TransferofData
Tier 3:
Analytics on
the Cloud
Tier 1:
Data
Sourcing
Tier 2:
Analytics on
the “Edge”
Figure 2
• High data volumes: Huge amounts of data in the cloud will spur scalability issues and pose a
problem of bandwidth, increasing storage costs. This factor makes it the primary business driver
of edge analytics.
• Latency: This can be a killer where predictions drive decisions and actions in real time. It may not
be prudent to have predictions made in the cloud and then sent back to ground zero. In the IoT
world, prediction has to be made within 100 milliseconds, which can be accomplished only with
edge analytics.

8
• Robotics: Autonomous systems, which are often used to inform decisions such as object recog-
nition, shortest path and control of actuators, must be made in real time. Edge analytics is an
essential component in these deployments.
• Power saving: Continuous data transmission to the cloud leads to huge power consumption.
Avoiding the transmission of data is one way to conserve power in low-power IoT devices.
• Regulatory problems: Countries such as Germany have passed regulations to keep the data
generated in their country within their political boundaries.2
Cloud servers are often located thou-
sands of miles away from the data’s originating country. With such new regulations, taking the data
to the cloud for analytics will become an issue. Edge analytics again will come in handy in such
situations.
ADDRESSING THE REQUIREMENTS
In the digital age, data has become the most valuable asset to organizations. Data-driven companies
are heavily dependent on sophisticated and heavy-duty analytical techniques to achieve business
objectives and stay competitive with rivals, globally. And edge analytics, without proper implementa-
tion know-how, will not add any value to envisioned business outcomes. Hence it is imperative that the
aforementioned business drivers are properly addressed to advance corporate goals.
High Volume Data
As the pressure increases on CIOs to reduce the volume of data transferred to the cloud, there is a
growing need for decentralized distributed computational power at the edge of the networks, close to
the generation sources, to monitor data volumes. Monitoring this high volume of data at the edge can
deliver meaningful insights in the following ways:
• Threshold crossing alerts (TCAs): The majority of the data received at the edge may not be
of much interest, assuming the system is working normally. To filter out such data, companies
can install a tool or software with predefined threshold values for parameters. When the param-
eter value crosses the threshold value, it will trigger an alert to the monitoring tool. Using this
approach, companies can save a lot of time and money when evaluating this data.
• Summary extraction: With this technique, companies can extract a summary of the analyzed data
at the network’s edge. This summary can then be sent to the cloud for more sophisticated analysis.
Companies can set a timer to extract the summary on a periodic basis. Doing so can dramatically
reduce the amount of data being sent to the cloud.
• Parameterized models: This more sophisticated technique is an advanced version of summary
extraction. Here, companies run an appropriate algorithm on the data periodically and extract only
the parameters for the model and not the complete summary. This method is used with advanced
computing techniques to transfer only the parameters to the cloud. In the example below, only
after the model is extracted in the cloud are the parameters used in the next iteration onwards.
This limits the amount of data sent to the cloud.
»» Parameterized model: a𝑥 +b𝑥2
+c𝑥3
+d, Extracted Parameters (coefficients): a:5, b:8, c:3, d:9.

QUICK TAKE
The Case for Connected Energy
Large corporate office complexes and buildings benefit from hardware/soft-
ware tools that monitor energy generation and consumption in real time.
The data is typically collected from all energy sources onto one IoT device
at the so-called edge of the corporate network. On this platform, the data
(kilowatt units) is continuously monitored around a predefined threshold
limit. When the power generated falls below this threshold, an alarm is acti-
vated. In this technique, data is not always sent to the central servers – only
when an abnormal situation occurs.
This real-time monitoring technique helps corporations maximize their use
of renewable energy. It saves bandwidth and space storage in the cloud. It
provides environmental benefits by helping companies reduce their carbon
footprints and meet regulatory targets on energy consumption. By prevent-
ing energy waste, corporations can monetize unused power by selling it to
the grid after work hours and during the weekends.

10
Low Latency Decisions
Edge analytics simulates the working of the
human nervous system: The edge environment
simulates the spinal cord and the cloud simulates
the human brain. Most of the sensory signals that
require instant reflex action are responded to
directly by the spinal cord without transmitting
the signals to the brain. Similarly, data requests
that demand immediate action are analyzed on
the network edge and are responded to immedi-
ately. This swift analysis drastically reduces the
latency overhead that results when data is sent
to the cloud for analysis.
The spinal cord transmits only those signals to
the brain that do not require an impulsive reflex
action, or those signals that can be handled only
by the brain. Similarly, the edge filters the data
that is sent to the cloud. The results of the edge
analysis that require more heavy-duty analytical
treatment are routed to the cloud through the
gateways at the edge of the network.
In our Digital Technologies Lab, we have built
solutions using edge analytics that have reduced
decision-making time to a few hundred millisec-
onds, from a few seconds.
Low latency can be achieved using two techniques:
• Static reflex: Reflex action at the edge is
triggered by the prediction made by the
computing model. This model can be a static
version, unchanging over time. A static model
is used when the data environment does not
change much and the corresponding data
falls into fixed patterns. The upside of this
approach is the model needs to be refreshed
only infrequently.
• Adaptive reflex: Adaptive models are more
sophisticated, and learn as they evolve. Such
adaptive models change and enhance their
analyzing capabilities according to the data
set. These adaptive models are used to handle
more complex analysis where data changes
are quite frequent.
Comparing the Human Nervous
System & Edge Computing
!
CLOUD
EDGE
Data
Sources
Gateways
&
Sensors
Figure 3

QUICK TAKE
Indoor Navigation
Indoor navigation tools provide navigational services inside a building when
and where GPS is not available. The objective is to provide the shortest path
between two points as people move from place to place.
This technology displays real-time locations, similar to how GPS works. Since
the technology operates with minimal latency, a model is trained using pre-
dictive algorithms that run on a server at the network’s edge, and which
are then distributed to the user’s mobile phone during runtime (when the
person is moving). Indoor navigation functions primarily on a hard-coded
reflex action since the models are pre-configured and changed infrequently.
In-Place Analytics
In-place analytics tools/software are designed to analyze and process data in its local environment.
By employing in-place analytics on native data, organizations can filter down these large data sets
collected from their IoT devices. The result generated from the analysis is more targeted and helps
organizations gain quick access to key facts about operational matters, thus empowering them to
make more informed strategic decisions. Working with smaller data collections also helps organiza-
tions avoid duplication of data and to more efficiently dispose of data that isn’t business critical.

12
One of the approaches to implement in-place analytics is geo-distributed machine learning (GDML)
(see Figure 4).
• GDML: In the GDML technique, each compute node runs a local component of the algorithm on
the data available and calculates interim results — local values of objective function, gradient and
direction.
»» The interim results are communicated to a central node on the cloud layer.
»» At the central node, a global component of the algorithm aggregates all the results it receives
from the network’s edges and calculates global approximate objective function, gradient and
direction.
»» Global approximate values are broadcast back to the edge layer.
»» This iteration is repeated until a desired convergence in the model is achieved.
THE ROAD AHEAD FOR BUSINESSES
Edge analytics provides an appropriate platform for numerous IoT services and applications, such
as driverless cars, smart grids/buildings and smart cities, as well as wireless sensors and actuators
networks (WSANs). Every business opportunity across any industry that requires low latency and
communications accuracy — including automotive, consumer electronics, energy and utilities, and
healthcare — will find the implementation of edge analytics extremely helpful.
Three-Tier Edge Analytics Architecture
Sensors
Sensors
Edge Layer Cloud Layer
Training
Temporary Model
Edge Layer
Gateway
Prediction
Action
Training
Model Convergence
Gateway
Prediction
Action
Training
Model Convergence
Data Data
Figure 4

QUICK TAKE
The Case for Smart Cities
In a smart city, traffic lights at each intersection are equipped with various
sensors to monitor passing cars and predict whether a driver will jump the
red signal. To develop this model, the data collected from all the traffic sig-
nals must be analyzed in order to identify the patterns. This can be done by
collecting the data from all the traffic signals at a centralized server, or we
can use the GDML approach to develop the model.
In GDML, the data collected in each traffic signal is kept locally — only the
objective functions generated locally are sent to the central server. Once a
converged model is arrived at, it is distributed to all traffic signals, and it
is then used for prediction. This approach drastically reduces data trans-
mission rates and at the same time generates an effective and predictive
global model.
An enormous number of applications are available for edge analytics, and with IoT gaining force in
the coming years, businesses interacting directly with consumers will have to realign their business
models to earn strategic advantage over their competitors. These businesses should know that it
would become extremely difficult to build a growth strategy based on the existing legacy systems as
they are very expensive to maintain and have low response times.
In the future, cloud-only computing will continue to be costly and time-consuming. The expected
growth in IoT-based applications will encourage vendors to commercialize edge analytics to deliver
on the promises of lower latency, near-real-time response rates and more-optimized user experience
than existing cloud analytics systems can provide.

14
FOOTNOTES
1 https://which-50.com/iot-connected-devices-reach-20-4-billion-2020-says-gartner/
2 https://reef.apache.org/papers/2015-12-GeoML.pdf
REFERENCES
• www.travancoreanalytics.com/can-edge-analytics-future-big-data/.
• http://analyticsindiamag.com/edge-analytics-taking-data-processing-from-cloud-to-edge-of-network/.
• http://ijarcet.org/wp-content/uploads/IJARCET-VOL-2-ISSUE-2-568-571.pdf.
• http://ijcsmc.com/docs/papers/October2014/V3I10201441.pdf.
• www.cio.com/article/3200846/cloud-computing/the-difference-between-edge-and-cloud-computing-all-cio-s-should-know.html.
• www.ioti.com/iot-strategy/why-iot-ecosystem-nervous-system.
• https://reef.apache.org/papers/2015-12-GeoML.pdf.
• www.gartner.com/doc/3675917/cool-vendors-iot-edge-computing.
• http://dataconomy.com/2016/03/fog-computing-future-cloud/.
• https://blogs.cisco.com/perspectives/iot-from-cloud-to-fog-computing.

Digital Systems Technology
Ranga Vangipuram
Chief Architect, Cognizant
Global Technology Office
Ranga Vangipuram is a Chief Architect within the Cognizant Global
Technology Office’s Digital Technologies Lab. He has 30 years of
experience in the manufacturing, telecomm and IT industries, with
nine of those years spent in the U.S. telecom sector. Ranga previ-
ously led new product development and innovation in Cognizant’s
social, mobile, analytics and cloud (SMAC) group. He is passionate
about emerging technologies and specializes in providing tech-
nological solutions to business problems. Ranga’s current area
of focus is in IoT and data science, with a focus on the connected
energy, indoor navigation, edge analytics and location-aware-things
spaces. He has a master’s degree in engineering from Anna Univer-
sity. Ranga can be reached at Ranga.Vangipuram@cognizant.com |
www.linkedin.com/in/ranga-vangipuram-8b2b541/.
ABOUT THE AUTHORS
Ashish Anand
Senior Business Analyst,
Cognizant Global Technology
Office
Ashish Anand is a Senior Business Analyst within Cognizant’s
GlobalTechnologyOffice(GTO).HeispartoftheGTOmarketingand
alliances team. Ashish’s passion is creating fresh perspectives on
future technologies. He received a postgraduate degree in business
management from the Institute of Management Technology, Gha-
ziabad. Ashish can be reached at Ashish.Anand6@cognizant.com |
www.linkedin.com/in/ashish-anand-265724102/.

GLOBAL TECHNOLOGY OFFICE
Global Technology Office (GTO) is a core business unit of Cognizant, with a mandate to power technology-driven transformation. GTO’s
mission is to power and accelerate Cognizant’s capability to harness transformative technologies and enable our customers, people and
processes to navigate the shift in the work ahead.
As part of GTO, Cognizant Technology Labs is the specialist group that researches, pilots and prototypes emerging technologies with the
greatest potential to spark transformative business innovation. Our labs focus on pursuing tomorrow’s technologies to put our clients
ahead, and extending our footprint in specific areas where we believe technology trends are headed.
We develop solutions that bring real value to our clients’ businesses, and help them own and embrace emerging technologies in ways that
make sense for their organizations. We also draw on our existing building blocks and emerging technology expertise to accelerate the devel-
opment of new applications.
ABOUT COGNIZANT
Cognizant (NASDAQ-100: CTSH) is one of the world’s leading professional services companies, transforming clients’ business, operating and
technology models for the digital era. Our unique industry-based, consultative approach helps clients envision, build and run more innova-
tive and efficient businesses. Headquartered in the U.S., Cognizant is ranked 205 on the Fortune 500 and is consistently listed among the
most admired companies in the world. Learn how Cognizant helps clients lead with digital at www.cognizant.com or follow us @Cognizant.
© Copyright 2017, Cognizant. All rights reserved. No part of this document may be reproduced, stored in a retrieval system, transmitted in any form or by any means,electronic,
mechanical, photocopying, recording, or otherwise, without the express written permission from Cognizant. The information contained herein is subject to change without notice. All
other trademarks mentioned herein are the property of their respective owners.
Codex 2938
World Headquarters
500 Frank W. Burr Blvd.
Teaneck, NJ 07666 USA
Phone: +1 201 801 0233
Fax: +1 201 801 0243
Toll Free: +1 888 937 3277
European Headquarters
1 Kingdom Street
Paddington Central
London W2 6BD England
Phone: +44 (0) 20 7297 7600
Fax: +44 (0) 20 7121 0102
India Operations Headquarters
#5/535 Old Mahabalipuram Road
Okkiyam Pettai, Thoraipakkam
Chennai, 600 096 India
Phone: +91 (0) 44 4209 6000
Fax: +91 (0) 44 4209 6060

Making Actionable Decisions at the Network's Edge

More Related Content

What's hot (20)

Similar to Making Actionable Decisions at the Network's Edge (20)

More from Cognizant (20)

Making Actionable Decisions at the Network's Edge