IEEE Consumer Electronics Magazine - May/June 2023 - 99

reality (VR/AR), video surveillance, and autonomous
vehicle) with requirements that cannot be
met by this centralized cloud architecture. For
example, VR/AR applications require millisecond
response time and thus cannot tolerate remote
cloud service latency. Moreover, the transmission
of huge amount of IoT data to the cloud will
consume excessive bandwidth. Other concerns
include security risk and regulatory compliance.
These limitations have created an essential
demand for the next-generation cloud computing
that takes place closer to connected devices. Various
related computing paradigms have been proposed,
such as cloudlet, fog computing, and
multiaccess edge computing (MEC).1 While the
exact distinction between these paradigms
remains an ongoing discussion in the research
and standardization communities, they all represent
initial steps toward the decentralization of
cloud computing.2 They are generally referred to
as " edge computing " whose main purpose is to
deploy resources for data processing and storage
at the network edge (e.g., base stations). However,
it is remarkable that edge computing is not a
replacement but a complement to cloud computing.
Moreover, as edge computing is deployed
across locations, how to operate and manage
these edge deployments as a single cloud from a
single control plane to achieve operational consistency
is an emerging need. Therefore, a new technology
trend is shifting from edge computing to a
wider concept of distributed cloud computing,
aiming at extending the footprint for the cloud.3
The evolution toward distributed cloud is accelerated
by the development of 5G networks. Inorder
to process the ever-increasing volume of data traffic
more cost efficiently and enable novel services,
the entire 5G network infrastructure is currently
constructed based on the cloudification from radio
access network (RAN) to core network.4 Specifically,
network softwarization techniques, including
software-defined networking (SDN), network function
virtualization, and network slicing, enable multiple
network slices to be created as chains of
virtual network functions and provisioned according
to cloud service models. As such, 5G-oriented
distributed cloud is where mobile networking and
cloud/edge computing converge.4
Ahandful of works have fostered the concept of
distributed cloud; however, their perspectives are
May/June 2023
fragmentary and a clear picture of distributed
cloud is still hesitated. This article revisits and
elaborates a mature view of this emerging paradigm.
First, we present a precise interpretation of
distributed cloud. Second, we describe a typical
three-layer architecture of distributed cloud and
its enabling technologies. Third, we build a threelayer
distributed cloud prototype and evaluate the
performance of a service discovery/deployment
protocol based on hierarchical SDN and contentcentric
networking (CCN) architecture in terms of
service discovery time. Finally, we identify some
open challenges for realizing distributed cloud.
CONCEPT OF DISTRIBUTED CLOUD
Distributed cloud is an emerging concept. At
first glance, distributed cloud could be thought of
as a cloud model where a cloud service provider
(CSP) has multiple geographically distributed
data centers that allow customers to utilize cloud
services (CS) from any of them.5 However, the
most mature view of distributed cloud is the distribution
of public CS to different physical locations
outside the CSP's data centers, but which
are still centrally managed by the originating
CSP.3 It is the first cloud model that incorporates
physical location of CS as part of its definition. Historically,
location has not been relevant to cloud
definitions, and the cloud is referred to as centralized
services running in the CSP's data centers.
Whereas, distributed cloud deployments allow
customers to run their applications or individual
application components in different locations,
including public clouds, on-premises data centers,
and other network edge locations, tomeet specific
requirements for response time and regulatory
compliance. The originating CSP remains central
responsibility for operation and maintenance of
the distributed cloud infrastructure (DCI), while
cloud customers can access the CS wherever they
are located and manage them as a single cloud
froma single control plane.
ARCHITECTURE
Generally, a distributed cloud has a hierarchical
network topology.6 Figure 1 illustrates a threelayer
distributed cloud architecture consisting of
core cloud (CC), regional cloud (RC), and edge
cloud (EC). We envision a scenario of the
99
IEEE Consumer Electronics Magazine - May/June 2023

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