HP Innovation Journal Issue 12: Summer 2019 | Page 60
Cloud service providers have
struggled with this issue in
particular: finding a path to ingest
their customers’ legacy data into
their cloud services. One approach
has been to physically transport
arrays of storage with data.
Amazon, for example, has an offering called Snowmobile
consisting of 10 semi-trailer trucks packaged with 100 peta-
bytes (PB) of disks over a period of six months to ingest just
one exabyte of data. 5 While effective for legacy data, these
types of solutions fail for applications requiring real-time
analytics using the device-generated data at the edge.
THE IMPORTANCE OF EDGE COMPUTE
Thirty percent of all data generated at the edge is hyper-
critical or critical, meaning that failure to perform
real-time analytics, primarily at the edge, puts lives
and wellbeing at risk.
The combination of trapped data
and time-sensitive analytics is
moving more compute from the
cloud to the edge. This trend is
drawing investment and competi-
tors from enterprises and startups.
Amazon and Microsoft have created distributed software
versions of their cloud offerings (AWS and Azure, respec-
tively), and they are licensing these versions to partners.
They are also driving hardware reference designs and
selling full solutions to drive standardized edge device
architectures, much as Intel did with CPUs in the Per-
sonal Computer business.
The industry’s challenge is to do this sustainably. Endpoint
devices able to perform analytics on data today are too
energy-intensive to deploy at scale. For example, a typical
AI development workstation used by HP Labs consumes 1.4
times the energy of an average U.S. home. The analytics
engine used to develop fully automated vehicles consumes
2.5 times the energy of a typical U.S. home (i.e., 2,500
watts). IBM’s Watson computer, which beat reigning Jeop-
ardy champion Ken Jennings, used 4,000 times as much
energy (80,000 watts) as Ken Jennings’ brain (20 watts).
For these edge solutions to scale, devices performing
analytics must consume orders-of-magnitude less energy.
HOWEVER, CORE ANALYTICS ENGINES
MUST BECOME MUCH
MORE ENERGY EFFICIENT
New Energy Efficient Analytics Engines
are Needed to Provide Data Insight
An advanced AI development workstation uses
1.4X the energy of the average U.S. home.
IBM’s Watson beat the reigning Jeopardy
champion, Ken Jennings, using 80,000 watts
in the process versus ~20 watts for the human
brain (4,000x times the energy).
Autonomous vehicle prototype use around
2.5x the energy of the average US home.
“To put a system into a combustion-engined
car doesn’t make any sense, because the
fuel consumption will go up tremendously.”
—Wilco Stark, Mercedes-Benz’s
Vice President of Strategy
F See the “Compute Efficiency” article in this issue. P.63
CLOUD COMPUTE FOOTPRINTS MOVING TO THE DATA
Many Approaches to Edge Standardization and Analytics, With No Clear Winner
AZURE SPHERE
• IoT device reference design
• Requires Azure Cloud subscription
• Orchestration via Azure IoT Edge
58
HP Innovation Journal Issue 12
SNOWBALL EDGE
• Terabyte-scale local storage compute
• Local footprint of Amazon EC2
• Orchestration via AWS Greengrass
ANALYTICS APPLIANCE
• On-premise IoT analytics
• Hosts Microsoft Azure Stack
• Partner with Amazon