IEEE - Aerospace and Electronic Systems - December 2023 - 17

Mafakheri et al.
virtualization infrastructures functional block within the
MEC host through the Mm7 interface, which is between
applications and the MMSC. In CANARIA, the lightweight
Kubernetes (k8s) distribution k3sa is used as
ECTM, enabling it to perform configuration management
of containers, scheduling, service discovery of pods, and
management of replication. Unlike typical data center
cloud infrastructure, the hardware forming the cabin edgecloud
comprises cabin servers and seatback screens,
resulting in significant differences in
1) performance (CPU and storage);
2) CPU architecture (hybrid cluster x86, ARM);
3) network topology (in-cabin cabling versus top-ofrack
switches).
These special properties of the cabin edge-cloud advocate
a suitable version ofKubernetes for cloud-like management
and orchestration. k3s fulfill these requirements, as it
requires significantly fewer resources than the full k8s distribution
and is available for hybrid architecture clusters.
NETWORK CONNECTIVITY
The network connectivity layer is responsible for providing
connections between the living applications within the
MEC host, the local data network (LDN) or the data network
(DN) (i.e., Internet). In CANARIA, the two main network
connectivity options are wireless fidelity (WiFi) and
light fidelity (LiFi). For LiFi, an access point is installed in
the aircraft cabin and connected through the Power over
Internet (PoE)+ interface to the MEC host. The aggregation
of the two technologies could meet the requirements of an
ultradense environment inside the aircraft. Furthermore, to
increase the radio access network (RAN) capacity, the multipath
transmission control protocol (MPTCP) is deployed
([13]) on both servers and clients (seatback screens).
MPTCP makes use of multiple linked TCP connections,
i.e., WiFi and LiFi, to carry a single data stream.
NETWORK EDGE MANAGEMENT INTERFACE
The network edge management interface (NEMI) is a software-defined
distributed data exchange and edge management
framework that is composed ofa number ofdistributed
microservices that provide reliable edge-based data sovereignty,
provenance, curation, and system actuation. The
NEMI framework separates the application (control) and
information traffic (data) into different planes affording the
development ofuse-case-driven key feature modules such as
1) cognition (autonomous control-loop);
2) autodiscovery (edge registration);
ahttps://docs.k3s.io/
DECEMBER 2023
3) resilience (back-pressure);
4) reliability (guaranteed delivery);
5) data provenance; and
6) scalability (multiple data streams)
to new and legacy data pipeline and message broker solutions.
Thus, NEMI through its modular approach supports
both greenfield and brownfield deployments by establishing
a distributed data-sharing substrate that integrates and
facilitates the monitoring, visualization, and control of
federated edge infrastructures.
The NEMI architecture illustrated in Figure 2 consists
of four distinct components split across the edge nodes
and a central location in a star topology configuration:
1) Piper (data plumbing);
2) Savant (cognition/AI);
3) Gateway (API load-balancer); and
4) Panelo (visualization and operational dashboard).
In the NEMI context, the edge nodes are considered as distinct
active systems for data harvesting and control actuation
point of view, even though the edge nodes may have
peer-to-peer functionalities between them. In the NEMI
landscape, an active system is a closed system or infrastructure
arrangement that allows data harvesting and control
actuation.
The Piper serves as the common data plumbing and
autodiscovery module between the edge and central. It
facilitates the discovery and registration of an edge node
with the central unit during the bootstrapping of the edge
active system using a combination of heartbeats and timeouts.
Piper microservices provide the application programming
interface (API) to control the data flow, orchestrate,
and coordinate the harvesting of measurements (raw metrics
and aggregated data), and logs (syslog messages) from
the edge active system. Piper allows the harvested data
from an active system to be stored both on the edge node
(edge data lake) and central side (central data lake), either
as raw or aggregated data, in a dynamic configurable manner
depending on the use-case and data flow instructions.
The Savant module provides cognition and AI functionalities.
Thus, the implementation and technology stack
of a Savant module are highly use-case dependent. In general,
it is envisaged that a Savant module would provide
advanced cognition and AI features in the form of a compute
engine based on an ML platform with AI algorithm
and model [ML, deep learning (DL)] services. However, a
Savant module can also be as simple as a rule-engine or
an if-else-then decision tree. Savants consume the data
harvested by the Piper and provide actuation commands to
the active system to complete the closed control loop. The
Edge Savant is part ofa reactive control loop that responds
IEEE A&E SYSTEMS MAGAZINE
17
IEEE - Aerospace and Electronic Systems - December 2023

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - December 2023
