From acting on real time air
quality measurements to lighting,
building automation systems
(BAS) are inching closer and
closer to the edge.
March/April 2019 I 33
of data that is continuously streamed across
all the different sensors that measure these loads.
Everything from human activity to solar radiation
and precipitation can affect these loads and fluctuate
throughout the day, such as when the building is
too cold in the morning but too hot in the afternoon.
Think of the occupancy rate compared to temperature,
humidity and precipitation levels. Not only can
the systems be tuned in real time, but they can start
predicting and detecting anomalies, which might
not show up in a system until something fails. HVAC
is truly expensive and by adopting these machine
learning approaches, HVAC energy consumption
and demand can be greatly reduced.
A FEW BIG PROBLEMS
LATENCY
Latency is a big problem. Latency is how fast
one computer can talk to another spread out over
a physical distance, and it is a problem because
if that distance is too far then decisions cannot
be made fast enough. For instance, consider
self-driving cars. There is software that can now tell
the difference between a child running in front of
a car versus a dog. Nobody wants to hit either, but
asking the cloud to make that decision is too slow.
The software onboard the car needs to make a decision
immediately about what to do— slow down, slam on
the brakes, veer sharp left, is the car running into
another car? Latency becomes a big problem even
though the benefits of applying a lot of this software
are clearly valuable. Tesla cars have GPUs in them
and run Linux, but what about the semi-trucks that
may eventually replace 3.5 million workers in the
U.S.? No one wants an 18-wheeler barreling down the
highway in California waiting to get a decision from
some data center in Virginia on when it decides it
wants the truck to change lanes. However, latency
is not the only problem. Bandwidth throughput is also
a problem when streaming high definition video from
a few thousand cameras in an IP surveillance or smart
city application. Being able to analyze information in
real time presents challenges if the intent is to offload
to software in the cloud.
SECURITY
Data security is one of those industries that seems
forever locked in the dark ages as it has not only
remained relatively the same over the past 25-30 years
since the Internet, but it has arguably gotten worse.
One might question how to secure edge compute
deployments when it is known that traditional IoT
and connected devices have such a horrible security rap,
and not even Google or Facebook are immune to the
whimsical desires of hackers. Uber was recently sued
by every single state’s attorney general in the U.S.
including the District of Columbia for covering up
a data breach. The company is paying $148 million
in the national settlement; it originally thought it would
only cost them approximately $100 thousand. Facebook
is looking at a $1.6 billion fine for a recent data breach.
Google shut down its social platform because its security
was incredibly broken, thereby exposing private data
in as many as 500,000 accounts; it is now confronting
possible legal repercussions. The emperor has no clothes.
Traditional IoT and connected devices have routinely
been used to build the largest botnets. Unsecured
printers, routers, and IP connected cameras have been
mass “rooted,” which is hacker parlance for completely
taking over and using them as giant cannons against
unsuspecting websites to knock them offline, sometimes
for days at a time. There is a very large chance that
companies and enterprises have connected devices in
one or more of their buildings that are a part of a botnet.