A Knowledge Graph Driven Approach for Edge Analytics
type, application type and the capabilities
and data sources which connect them, lays
the foundation for the knowledge graph. The
instances associated with these application
and device types are also represented in the
graph to describe the current ecosystem
within the represented network. The
knowledge graph approach utilizes a
resource description framework (RDF) graph
paradigm to represent relationships and
connections between two nodes in a
knowledge graph instance as triples with a
subject, predicate and object (e.g. [Node_1,
"has Component", Node_2] with the
relationship expressed as "type Node_1 has
a component consisting of type Node_2").
The entire knowledge graph instance
demonstrates a series of "triples" from the
schema. A triple is a schema linkage between
a pair of nodes where the predicate is
effectively a type of edge and the subject
and object pair are the equivalent of node
types in the graph. This structural
representation allows for searching and
inferencing of information within a graph to
understand and reason upon implications.
Edge applications and the edge orchestrator,
a layer in the architecture housing several
services for facilitating communication back
and forth with the up-stream knowledge
graph, use this ontology to determine how
to interact with each edge device and what
supporting container infrastructure must be
launched.
applications communicate. While a
messaging protocol is chosen and fixed
during framework onboarding, the schema
of the data being transmitted is stored in the
knowledge graph. This backbone combined
with the information in the knowledge graph
allows for the decoupling of the different
components.
Containerization
All components on the edge appliance (from
the edge applications to the orchestrator
and the messaging backbone) are
containerized 6 . This enables the deployment
of the core components and the custom
applications on any hardware infrastructure
or OS. Additionally, it enables applications to
be dynamically deployed and moved, and
their resource utilization to be monitored
and controlled.
Edge Framework Domain Knowledge Roles
Revisiting our previous enumerated
scenarios, our edge analytics framework
would enable each domain-knowledge
expert to onboard, develop applications for,
and deploy a new type of sensor in the roles
as follows:
Device Expert
To onboard a new type of sensor, the device
expert is responsible for updating the
knowledge graph with the relevant
information regarding a new sensor type,
namely, the upstream data message schema
streamed out of the sensor, the upstream
control message schema generated by the
sensor (e.g., heartbeat or status messages),
Messaging Backbone
An
inter-component
communication
interface through which all edge devices and
6
“What is a Container”, 2017. [Online]. Available: https://www.docker.com/what-container.
IIC Journal of Innovation
- 49 -