IEEE Consumer Electronics Magazine - May/June 2023 - 59

Cloud environment, composed of static, largescale
data-centers (see the section " Differences
Between Cloud and Edge Infrastructure " ). Moreover,
Cloud applications are usually packaged
either as virtual machines or containers. While
these virtualization technologies provide isolation
for applications sharing the same infrastructure,
they introduce a limitation on the features
they can access on the host machine. Far-Edge
applications in use-cases such as Edge robotics,
UAVs, and vehicles may require direct access to
sensors and actuators that, for security reasons,
should be given via privileged containers/VMs.
This is an example of new characteristics
introduced by the far-Edge concept. These
characteristics are very different from the
ones that can be found in a static data-center;
as a consequence, they are not considered in
current management solutions. Therefore,
extensions to the current information model
should be considered in order to: (i) keep
track of the different capabilities and mobility
characteristics of underlying infrastructure
and (ii) allow broader application packaging
technologies (e.g., robotic applications, microcontroller
applications).
The next issue identified deals with the consistency
of the systemand its degree of centralization.
Before discussing them, it is worth introducing the
CAP theoremand its implications in both the Cloud
and Edge domains. The CAP (i.e., consistency,
availability and partition) theorem was introduced
in Brewer et al.,17 and it states that any distributed
system that has to deal with data can only provide
twoof the three following characteristics.
Consistency: Everyone who reads information
from the system gets either the most recent
information or an error.
Availability: Everyone who reads information
gets a (nonerror) response without the guarantee
that this is the most recent one.
Partition tolerance: The system continues its
operations despite messages being dropped
by the network connecting the system
elements.
Current Cloud solutions cope with the problem
described by this theorem by leveraging on
high-redundant networking and replication of
May/June 2023
data and services (see Figure 1). This approach
allows us to reduce at minimum the risk of partitioning,
and therefore, such systems can operate
by choosing Consistency and Availability as
their guarantees, with the Partitioning provided
by the underlying physical infrastructure. However,
in the far-Edge environment, it is unfeasible
to reach such level of physical redundancy,
resulting in a high risk of system partitioning
that must be considered. To get a tradeoff
between the three guarantees in the far-Edge
scenario, it is needed to explicitly handle partitioning.18
This need goes against the centralized
control and systems state considered in Cloud
environments.
Centralized control plane: As illustrated in
Figure 2, each of the analyzed solutions is based
on a centralized control plane. Such design is
common in Cloud solutions (see the section
" Cloud and Edge Management Solutions " )
because it facilitates development. To achieve
high-availability and consistency, replication
techniques, inherited from state-of-the-art, web
approaches can be used. However, as stated
above, to achieve such properties in the farEdge,
it is necessary to explicit handle partitions.
This means that control functions such as
scheduling and monitoring need to be aware of
the possible partitioning of the system, and take
decisions involving the tradeoff between consistency
and availability.
In the light of this, we believe that distribution
of control functions and intelligence among
nodes in the system should be considered,
because (i) each node can intelligently react to
events even if it is partitioned and (ii) the whole
system does not stop its evolution in case of
partition.
Centralized system state: Current solutions are
based on a centralized storage for the whole system
state that acts as source of truth (see the
section " Cloud and Edge Management Solutions " ).
This approach is common in Cloud and
near-Edge, since they can maintain easily a centralize
repository of information. However, as
shown by experimental results from the section
" Experimental Evaluation, " this requires a good
amount of system memory devoted to the storage
of state information, yielding to the need of
powerful central servers to store and provide
59
IEEE Consumer Electronics Magazine - May/June 2023

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