IEEE Consumer Electronics Magazine - May/June 2023 - 58

Edge Computing
FIGURE 6. Average deployment factor - DFðx; yÞ¼ TDeploy½K8s½1;10
TDeploy½x;y
instantiation required in KubeEdge ranges from
26.1 to 40.4 times longer than Kubernetes and
from 18.1 to 21.1 times longer than K3s. In absolute
terms, this translates into tens of minutes
instantiation time with KubeEdge compared to
tens of seconds in Kubernetes and few minutes
in K3s. This is because, with a limited number of
nodes in the cluster, KubeEdge tends to allocate
all the containers in a subset of the available
nodes, resulting in the nodes being overloaded
and a longer overall service instantiation time.
Summarizing, from the presented results we
observed the following.
1) The way solution component are deployed
(containerized or simple binaries) influences
memory utilization.
2) Memory utilization is dominated by the number
of nodes composing the infrastructure.
3) Monitoring of the infrastructure requires
approximately two orders of magnitude
more bandwidth compared to container
monitoring.
4) Scheduling decisions may overload some
nodes increasing the overall service instantiation
time.
DISCUSSION
This work discusses the issues of today's centralized
management solutions when applied to
the far-Edge; an environment heavily characterized
by volatility, mobility, and decentralization
of resources. During our analysis, we have identified
several aspects to consider in the design of
the next generation Edge systems: Infrastructure
58
: (a) Kubernetes. (b) K3s. (c) KubeEdge.
monitoring. It plays a critical role in the management
systems where the life-cycle management
of the applications highly depends on the realtime
information about the state of the infrastructure.
To support this information, current
management systems (see the section " Cloud
and Edge Management Solutions " ) adopt a
stream-based monitoring approach, which poses
stringent requirements on networking, like
always-on connectivity, minimum required
bandwidth and low-latency. However, based on
our analysis of the far-Edge infrastructure in the
section " Differences Between Cloud and Edge
Infrastructure, " this scenario is characterized by
unreliable networking without any guarantees.
Hence, a stream-based approach can led to saturation
of the already scarce networking resources,
resulting in a harmful behavior. In addition,
the use of outdated information in real-time decisions,
e.g., scheduling, may yield to suboptimal
performance. A different approach is then
needed to overcome this issue like new monitoring
paradigms designed to work over low-bandwidth
and unreliable networks. Examples of
such paradigms are event-based and thresholdbased
monitoring, which relax requirements
over the network.
Information model is the representation of
concepts, relationships, constraints, rules, and
operations used to describe a certain domain. In
our case, it refers to how each solution describes
the underlying infrastructure, applications, and
their relationships. Based on our analysis in the
section " Cloud and Edge Management Solutions, "
current solutions are designed for the
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - May/June 2023

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - May/June 2023
