By Eric Broockman, CTO, Extreme Networks
In my position, I meet lots of great
technologists, from semiconductor practitioners to cloud application
developers, from IT staff of small cities to CTOs of some of our most
successful channel partners as well as execs from our largest customers.
Frequently I get asked, what is one of the newest technologies we should be
paying attention to? I often answer with - "Watch for the coming era of eyeballs
at the edge".
What does this really mean? It isn't about the
profusion of video cameras that continue to crop up in building or on campuses
all around our cities. Most of those are related to physical security. Rather,
the technology inflection point and trend I see hurtling towards the enterprise
and becoming more real every day is video analytics at the edge. So why call it
eyeballs at the edge? When you look at the various use cases, many of
them are an attempt to apply what by human standards are essentially relatively
simple visual, think eyeball, image recognition capabilities to a video stream
source coming from a digital video camera. What is especially interesting
however, is that the chips and other infrastructure that enable this capability
are undergoing an inflection point in price-performance, size, throughput, and
power consumption. What was thought of not long ago as a huge CPU hog, that
then became a super high-power specialty version of a GPU chip, is now becoming
a far more modest chip, or collection of chips, attached to PCI or USB
connections that cost less, are much smaller and consume far lower power.
But how is this eyeballs at the edge? Let's
cover a few examples. In healthcare, a surprising amount of time is spent by
very important staff simply looking for expensive portable medical equipment.
Eyeballs on the lookout. Think mobile X-ray machines or ultrasounds or infusion
pumps, etc. A great deal of money has been spent by institutions using various
forms of wireless technology to help find these high value assets. They help,
but their precision can be suspect. Though the future of UWB locationing holds
great promise, another way to look for stuff is to look for stuff with eyeballs
at the edge. In particular, using video analytics to identify important objects
as they move down hallways in health clinics - just like a human eyeball does.
Video analytics can be trained to easily recognize any number of important
high-value assets in a healthcare setting. This capability enables new asset
location applications to identify that the mobile X-ray machine that used to be
in room 217-B, traveled down the hallway and can now be found in room 263. Note
that this application is a true edge application for a few reasons. In some
instances, being able to give relatively immediate real time asset location
information precludes using a distant cloud to perform the video recognition
and analytics. In addition, the sheer volume of data, even if it is already
compressed, that would be sent to a cloud from 100s of video cameras in a
modern healthcare facility would generate an exorbitant cost in simply
transporting the bits to the cloud. Taken together, this drives the need for
edge computing. The newest technology coming to market makes this possibility
remarkably affordable versus just a few short years ago.
Perhaps another example. Today, in many
stadiums there are already people counting "eyeballs at the edge" applications,
keeping track of queues at bathrooms, etc which then trigger electronic signage
or alert venue engagement applications on your smartphone when the queue times
become uncomfortably long. A more sophisticated use case however is sentiment
analysis in a retail environment. If you were part of a retail digital
engagement team, wouldn't it be useful to be able to tell when a customer
standing in front of an item on display was interested, disinterested, curious,
needed help, etc? Video analytics can help here. The algorithms are getting
good enough to detect more than simple happiness or unhappiness. They can give
insights into whether a customer is looking for help - much like a human
associate can tell by simply watching the facial expressions of a customer.
"Can I help you? It appears you are a bit perplexed by this product demo."
What about a higher education example. Video
analytics are getting good enough to tell the difference between a group of
friends walking home from a game at the stadium on campus and celebrating, and
two people in the midst of a fist fight - or worse. What about license plate
recognition in cities to assist traffic improvement software in a smart city?
As the algorithms get more advanced, the true
power of eyeballs at the edge becomes ever more real, and the usefulness
greatly expands as well. There are certainly some Orwellian creepy use cases
that can be conceived of, but like any new technology, the potential for the
greater good is high. So why is this trend about to suddenly become a "thing"
you might ask. Like any inflection point, it is usually because of a step
function change in capability of some underlying technology. The step function
here has not been the change from X86CPU to an Nvidia GPU, though that was a
step that greatly accelerated the field. Rather, from my vantage point, it is
the newer inflection point of custom video AI accelerator chips from companies
like Intel with their Movidius chip, or Google and their small TPU chip, or new
startups like Mythic with their promised and forthcoming unique mixed digital
and analog approach. These chips enable the capabilities of far-superior
price/performance/watt solutions that may be delivered in an edge computing
server, or be part of a smaller form factor, ruggedized Intel NUC-like form factor
or the new video cameras coming to market with embedded video AI, and even
Razor's Edge computing models where the video accelerators are blades powered
by PoE directly attached to wiring closet switching systems. When these types
of step functions - inflection points - happen, following the ideation
gestation period, there is a plethora of new applications, which begin to grow
rapidly, cross the proverbial chasm, and become part of our mainstream way of
life. Eyeballs at the edge is one of these promising and forthcoming inflection
points to keep your eyes on.
##
About the Author
Eric
Broockman is the Chief Technology Officer at Extreme Networks. He develops and
drives the company's technical strategy, provides strategic insights and
direction on new technology inflection points, and leads technical due
diligence and integration planning for Extreme's acquisition strategy. In
addition, he evangelizes the company's technology vision and cloud leadership
in the enterprise networking market, differentiating Extreme's point of view
and technical brand.
Eric brings diverse career experience across
multiple fields, ranging from communications software, networking products,
retail systems, as well as networking and consumer semiconductor products. Most
recently, he was the CEO and founder of Alereon, a gigabit UWB wireless startup
company. He has held positions as VP/GM for networking businesses, as CEO for
microprocessor, networking, and wireless start-up companies, and as EVP of
Marketing for a networking semiconductor company. Eric has led advanced
networking product groups at IBM and Crystal Semiconductor and was a VP/GM and
executive officer of Cirrus Logic and Legerity. He holds a BS in Engineering
from the University of Florida, a MS from the University of Florida and is a
graduate of the UNC Chapel-Hill Executive Program in Business Administration.
While in graduate school, Eric was a National Science Foundation Fellow, holds
8 US patents, and has published 33 technical disclosures and multiple journaled
articles.