IEEE Consumer Electronics Magazine - May/June 2023 - 104

Distributed Cloud Computing: The Big Picture
Service Deployment and Discovery
Distributed cloud opens up the opportunity
for third-party SPs to deploy their services at
any ECs close to end users. Since SPs usually do
not know exactly where and when a service
should be deployed, it is necessary to develop
mechanisms for SPs to automatically deploy
services to any ECs when requested. Another
concern is how to route a service request to the
EC where the service is running. Since the network
location of service instances may change
at any time, distributed cloud requires a servicecentric
architecture where user applications can
request a cloud service without any information
about its location. A combination of CCN and
SDN technology offers promising solution.14 The
former enables service discovery based on service
names, while the latter enables location
tracking of services. A preliminary evaluation of
service deployment and discovery based on
SDN-CCN architecture was presented in the
" Case Study " section. In addition, AI/ML can be
applied to analyze dynamic service demand and
network conditions for improved performance
in large-scale deployments.
High Mobility
Distributed cloud is expected to support CS
with high mobility and low latency, which means
ensuring the service continuity and quality of service
(QoS) as the user moves across different network
coverage areas. To this end, seamless
service migration plays an essential role, determining
to which ECs the service should be
migrated, and howto efficiently transfer the user's
service state (e.g., memory state data, application
image data, and user settings) between multiple
ECs, considering the overhead and QoS requirements.
It is a challenging task due to the heterogeneity
and fluctuation of cloud and networking
resources, dynamic user mobility patterns, and
security and privacy concerns. Most recent studies
have leveraged SDN paradigms to implement
state transfer protocols and enhance mobility
management.8,9 However, efficient SDN multicontroller
coordination to support high mobility and
ultralow latency requirements of emerging applications
for large-scale deployments remains an
open issue. Mobility prediction via AI/ML and
slicing-based mobility management where an
104
optimized network slice is dedicated to mobility
functions are also directions for further research.
Load Balancing
The temporal and spatial diversity of application
demands may incur the problem of workload
imbalance and network traffic congestion
among distributed clouds. Some ECs may be
overloaded when handling a large number of service
requests from users due to resource constraints,
while others are in idle or underutilized
state. In this regard, load balancing is an important
research topic in distributed cloud, which
intends to distribute workload evenly among
cloud nodes and network domains for optimizing
the use of resources. Various load balancing
methods have been proposed and divided into
static and dynamic algorithms according to
whether the current state of computing nodes is
taken into account to make distribution decisions.
However, while some parameters like service
latency, resource utilization, energy
consumption, and operational cost are widely
considered for performance evaluation, other
parameters like scalability, reliability, and security
are mostly overlooked.13 Therefore, load balancing
decision making requires being extended
to bring more QoS parameters and the tradeoff
between them into consideration. On the other
hand, the coordination between distributed
clouds for load balancing based on decentralized
SDN architectures needs thorough evaluation
in future research.
Edge Intelligence (EI)
Distributed cloud lays a promising foundation
for EI, which is commonly considered to be
a key missing element of 5G and predicted to be
a major enabler for 6G.15 EI refers to a new technology
framework that pushes AI computations
(e.g., model training and inference) from the
cloud to the edge as much as possible to facilitate
various low-latency distributed intelligent
services. Due to the complexity of AI computations
and the heterogeneous and resource-constrained
edge environments, there are several
research areas to be explored toward realizing
EI. New distributed ML architectures and algorithms
need to be developed so that AI model
training and inference can be divided into
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
