SDN BASED LOAD DISTRIBUTION AND QOS ROUTING IN
UNDERWATER NETWORKS
INTRODUCTION within the networks. Due to these
applications’s different bandwidth and
In recent years, there has been an
medium requirements within the network,
increasing growth in smart cities. One of the
effective load balancing strategies and
most important technologies used in it is the
Quality of Service (QoS) management are
Internet of Things (IOT), which is defined as
essential for efficient and reliable
the “infrastructure of information society”
communication. To establish these, proper
[1]. IOT has enabled numerous technologies
development of network architecture and
on land, such as industrial automation,
intelligent routing protocol is required. In
healthcare, and smart homes [2], [3], while
previous years, most of the networks were
most of the underwater areas remain
developed with a closed-form architecture
unexplored [1]. This has led to the
that was difficult to reconfigure, reprogram,
development of the Internet of Underwater
and evolve as it involved manual
Things (IoUT). It is increasingly becoming a
configuration of switches and routers within
powerful technology for developing the
the network [12]. These limitations are
smart ocean [4]. It helps to manage the
overcome by the proposal of Software
increasing demand for the real-time
Defined Networks (SDN), which provides
collection of data and the management of
centralized solutions that can be used to
underwater sensors [5]. It is a network of
address these issues [13]. It allows the
smart sensors and devices that communicate
separation of control plane and data plane
with each other using the internet. They can
for delivering the most flexible and
also communicate with the systems that are
manageable networks. The SDN has three
present in the land and have evolved from
planes, namely, the application plane, the
Underwater Wireless Sensor Networks
control plane, and the data plane. The
(UWSN) for real-time data collection [5],
application plane is responsible for
monitoring environmental parameters, and
managing the logic of the applications that
producing scalable solutions. This is made
define the behavior of the network. The
possible through the active sensor nodes
control plane is the brain of the SDN
(i.e., Automated Underwater Vehicles
network [14]. It has centralized control over
(AUV) [6] and Remotely Operated Vehicles
the network and is responsible for defining
(ROV) [7] that can move around and
the way to handle the application. It also
interact with the environment and
acts as an intermediate between the
communicate data. IoUT facilitated many
application and the data plane, which
applications, which included subsea asset
translates high-level requirements from the
surveillance and management [8], military
application plane to low-level instructions
defense [9], video-assisted and image-based
for the data plane [15]. The data plane is
target tracking [10], [11].
responsible for executing the instructions
These applications have led to the and forwarding the data packets that contain
production of a huge amount of data that is the information.
of different types, generating data traffic
Though the centralized controller load controller to remove the extra workload
offers easy control, it also becomes a single [17]. Though the approach seems efficient, it
point of failure for the entire network [14], does not take the migration side effects, such
[15]. Due to this issue, the upcoming as migration conflict and high response
networks started using a semi-centralized time, into consideration. Hence, we intend to
control plane. The plane consists of multiple address these existing issues through an
controllers that are physically distributed but efficient controller allocation mechanism
logically centralized, thereby allowing along with a migration optimization model.
access to the distributed architecture’s As we discussed in paragraph 2, routing
scalability and preserving the centralized protocol is essential for performing QoS-
system’s simplicity [16]. Ensuring reliability aware routing. Traditional routing protocols
and scalability for this multiple controller used a cross-layer communication approach
architecture in SDN becomes critical as the [18], which makes it difficult to manage the
network scale enlarges. The dynamic network. The combination of SDN and
changes in traffic patterns lead to the uneven reinforcement learning has achieved greater
distribution of load in the network of success in the routing protocol [19].
multiple controllers, thereby making it However, achieving multiple performance
essential for the system to implement an metrics like latency, bandwidth, and energy
effective load balancing strategy. consumption is still a challenge. Hence, we
intend to address this issue using multi-
This issue can be handled by switch
objective reinforcement learning (MORL) to
migration, in which a whole switch is
achieve multiple objectives for efficient
transferred from a busy controller to a light
QoS-aware routing
.
