IEEE Consumer Electronics Magazine - May/June 2023 - 102

Distributed Cloud Computing: The Big Picture
complex orchestration of a vast number of containerized
services.
AI-Powered Intent-Based Networking (IBN)
The great potential of AI can find in distributed
cloud a breeding ground, not only to create
novel services but to aid in network orchestration.
With AI-driven automation, a new paradigm
called IBN is emerging, in which the users' input
to the cloud system is confined by the desired
business objectives or intents (e.g., service level
agreements).11 The service orchestration module
then decides how best to achieve those
intents, with the first step being to translate
users' intents into network policies. The second
step is to allocate resource at each domain and
automatically activate the codified policies to
provision the desired intents. Finally, the service
orchestration will continuously monitor and
take proactive actions for service assurance by
using machine learning (ML) to learn from historic
or real-time data. These steps form a
closed-loop automation solution with zero-touch
experience, so the complexity of service orchestration
can be hidden from customers.
SERVICE DEPLOYMENT AND
DISCOVERY IN DISTRIBUTED CLOUD:
A CASE STUDY
In this case study, we set up a three-layer distributed
cloud prototype, including one CC and
two RCs, each of which manages two ECs. The
CC is deployed in a VM on Amazon EC2. The RCs
and ECs are deployed on PCs in our laboratory.
The ECs are configured as WiFi access points
and receive service requests. We develop a protocol
to help third-party SPs self-deploy their
services in the ECs and assist the ECs self-discover
the requested service instances in neighboring
ECs. As shown in Figure 3, each cloud
node is implemented with three main components:
cloud service pool, SDNcontroller, and CCN
database. We use Docker container12 to deploy
CS. Each CS instance is implemented in one
container, and Docker manager has roles to initialize/start/stop
the containers. The service
deployment APIs includes a list of REST and
Python APIs for the frontend with SPs and the
102
FIGURE 3. Distributed cloud prototype.
backend with Docker manager, respectively, and
the CCN database is implemented by MySQL.
At first, if an SP wants to deploy a service in
the ECs, it has to register the service with the
CC. The CC creates service identity (ID) and then
broadcasts this information to all cloud nodes to
store in the registered service store. As an incoming
request arrives at an EC, its service aware
module checks if the request is for the registered
services. If not, the EC forward the request to the
network as normal packets. Otherwise, the EC
stores the request in pending interest table, and
checks if there is a service instance running on
itself (in the online service store table) or on
another EC (in the forwarding information base
table), it will automatically route the requested
packet to the destination by using local routing
table. Otherwise, it requests the service discovery
module to discover the running service instance.
In case of no found service, the CC will send a
message, including the service ID and the IP
address of the EC requesting the service, to the
SP to deploy the service at this EC.
As shown in Figure 4, the service discovery
protocol will sequentially check service information
in the EC, RC, and CC. We compare the performance
in terms of service discovery time
between this hierarchical architecture and the
centralized SDN-CCN architecture where the CC
FIGURE 4. Service discovery process.
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
