Edge Analytics – The Pros and Cons of Immediate, Local Insight

Edge Analytics – The Pros and Cons of Immediate, Local
Insight

The Internet of Things (IoT) brings businesses many benefits,
primary access to more data and better insights. However, most companies we
have spoken with are still largely puzzled by what to do
with their data. Whether they should store or discard enterprise data, and if
stored, what the best approach is to making that data a strategic asset for
their company.

Gartner
estimates that there will be 25 billion “things” connected to the Internet by
2020. The sheer size and speed of data collected when every device involved in
your business process is online, connected, and communicating can strain the
sturdiest network infrastructure. As such, despite
the widespread proliferation of sensors, the majority of IoT data, collected is
never analysed, which is tragic. Many
existing IoT platform solutions are painfully slow, expensive, and a drain on
resources, which makes analysing the rest
extremely difficult. Furthermore, in situations where timing is critical,
delays caused by bandwidth congestion or inefficiently routed data can cause
serious problems.   

The
key takeaway is that data is the most valuable asset for any company. So it
would be a shame to completely discard or let it lie dormant in an abandoned
data lake somewhere. It’s imperative that all data scientists tap into their
swelling pools of IoT data to make sense of the various endpoints of
information and help develop conclusions that will ultimately deliver business
outcomes.  I am totally against of
discarding data without processing.

In few
years there will be an additional 15 to 40 billion devices generating data from
the edge vs. what we have today[1].
That brings new challenges. Just imagine an infrastructure transferring this
data to data lakes and processing hubs to process? The load will continue to
rise exponentially over coming months and years, creating yet another problem of
stretching the limits of your infrastructure. The only benefit of this data
will come from analysis either it is traffic of “things” or surveillance cameras.
In time-critical situations, if we delay this analysis it might be “too late”. The delay could be due to many reasons like
limited network availability or overloaded central systems. 

A
relatively new approach to solving this issue is called “edge analytics”. It is
as simple as to say, perform analysis at the point where data is being generated (or analysing in real-time on site). The architectural design of
“things” should consider built-in analysis. For example, sensors built into a train
or stop lights that provide intelligent monitoring and management of traffic
should be powerful enough to raise the alarm
to nearby fire or police departments based on their analysis of the local surroundings.
Another good example is security cameras. To transmit the live video without
any change is pretty much useless. There are algorithms that can detect a change, and if a new image is possible to
generate from the previous image, they
will only send the changes. So these kinds of events make more sense to be
processed locally rather than sending them over the network for analysis. It is
very important to understand where edge analytics makes sense and if “devices”
do not support local processing, how we can architect a connected network to
make sense of data generated by sensors and devices at the nearest location.

Companies
like Cisco, Intel, and others are proponents of Edge computing, and they are promoting their gateways as Edge computing
devices. IBM Watson IoT, an IBM and Cisco project, is designed to offer
powerful analytics anywhere. Dell, a typical server hardware vendor, has
developed special devices (Dell Edge Gateway) to support analytics on Edge.
Dell has built a complete system, hardware,
and software, for analytics that allows an analytics model to be created in one location, or the cloud
and deployed to other parts of the ecosystem.

However,
there are some compromises that must be
considered with edge analytics. Only a subset of data is processed and analysed. The analysis results are transmitted over the network, which means they
are effectively discarding some of the raw data and potentially missing some
insights. The situation that arises here is the consideration of whether or not
the “loss” is bearable. Do we need the entire data set or is the result
generated by the analysis enough? What will be the impact of only using a
subset? There are no generalizations to be made here. An airplane system cannot
afford to miss any data, so all data should be transferred in order to detect any pattern that could lead
to an abnormality. However transferring data during flight is still not
convenient. So collecting data offline when the plane lands and Edge analytics
during flight is a better approach. The others where there is a fault tolerance
can accept that not everything can be analysed.
This is where we will have to learn by
experience as organizations begin to get involved in this new field of IoT
analytics and review the results.

Again,
data is valuable. All data should be analysed
to detect patterns and market analysis. 
Data-driven companies are making a lot more progress compare with
“digital laggards”. IoT edge analytics is an exciting space, and many big companies are
investing in it. An IDC FutureScape report for IoT notes that by 2018, 40
percent of IoT data will be stored, processed, analysed,
and acted upon where they are created and
before being transferred to the network[2].

Sources: