A Knowledge Graph Driven Approach for Edge Analytics
of the first edge device is as simple as the
instantiation and addition of the 1000th
edge device. We accomplish this by
employing a unique knowledge graph
approach, standardized communication
enabled with a distributed message queue
and containerization for consistent compute
and driver deployment.
Common Deployment Scenario
Consider the following common client site
deployment scenario: (1) A device expert
wants to onboard a new type of sensor for
use in an edge analytics deployment, (2) An
application expert wants to develop or
refactor an analytics application that
consumes the data generated by that sensor
and (3) A field engineer will deploy several of
these new sensors when steps (1) and (2) are
complete.
P ROBLEM S PACE
Many of the challenges facing the design of
edge analytics frameworks are well
understood 4 . However, our goal is not to
design a vertically integrated edge analytics
solution anew. Instead, we aim to create an
edge analytics framework that tackles the
following challenges: (1) Seamless and rapid
integration with a client's existing
heterogeneous
environment
(varied
application portfolio and varied hardware
infrastructure) and (2) Provide an
abstraction layer that enables domain-
specific experts to develop edge analytics
components that are decoupled from and
agnostic to the implementation details of
the underlying connected components.
An ad-hoc solution allows for deployment on
heterogeneous hardware and software
infrastructure. However, it requires all three
roles in the scenario to have intimate
knowledge of the entire edge framework
and requires close coordination of their
work. Tight integration and coupling of work
efforts runs counter to a scalable or
maintainable solution because it doesn't
allow for separation of component design,
development and workflow modularity.
Additionally, tight coupling of work tasks
may require numerous task handoffs for
quality control or modifications that are
difficult to streamline and minimize.
Vertically-integrated solutions 5 attempt to
address this problem by controlling every
aspect of the edge framework from the
hardware to the cloud. While an IoT
framework may be partially decoupled,
system engineers still require knowledge of
and experience in that solution's complex
development environment for each of the
three components of the scenario.
What follows is an explanation of a
generalized use case, two in-flight domain
specific use cases demonstrating the cross-
domain applicability of our Edge Framework,
their core and distilled challenges and
resolutions.
4
M. Patel and B. Naughton, "Mobile-edge computing introductory technical white paper," White Paper, Mobile-edge Computing
(MEC) industry initiative, September 2014.
5
“Foghorn Lighening," 2017. [Online]. Available: https://foghorn.io
IIC Journal of Innovation
- 45 -