Instrumentation & Measurement Magazine 25-7 - 34

to conventional approaches to see which approach performs
better in allocating resources under dynamic traffic loads. The
algorithm is capable of operating under the control of numerous
controllers.
A method for ElasticCon based on migration is proposed by
Y. Hu et al. [5]. The pool is gradually brought into equilibrium,
and the performance of the network is improved in response
to changing traffic conditions. On the other hand, the model
has significant overhead. The proposed approach searches for
the optimal switch for migrating load balancing jobs. The allocation
that comes about as a consequence helps to lessen the
need for controllers and keeps the performance of the network
from decreasing [8]. Additionally, the authors performed a
load balancing job for the network with multiple controllers.
By distributing control traffic among several already accessible
controllers, the proposed network is able to eliminate the
problem of oscillation.
In addition, other authors suggest using a server-based
load balancing method [9]. The model then distributes clients
among available servers. The authors' performance is
based on a round-robin technique and randomization, and
the migrating server is the smallest and lightest of the servers.
Acceptable network performance could be achieved with the
authors' model. Another model selects many customers that
have many different requirements so that the load can be balanced
[10]. In this application, a roulette wheel, single-point
crossover, and single-point mutation are used. The fitness
function is implemented using the coefficient-based server
model. The model was evaluated using random and roundrobin
methods for comparison.
In other research, the model is optimized to ensure that the
cloud consumer application is balanced [11]. Particle swarm
optimization (PSO) improves the application's average response
time, increases throughput, and optimizes resource
consumption. Dynamically loaded maps for traffic analysis
are available online. Pool-based server balancing is the method
used to achieve high performance. The researchers in [12] used
a very similar method to allocate requests to the server. The
main goal of the SDMN-based load balancer is to ensure that
applications receive an effective distribution of the resources
and tasks available to them. SDMN's ability to map distribution
dynamically. During the period when the network was
evolving dynamically, congestion-based multi-meter load balancing
was also implemented.
Methodology
In the following paragraphs, we will talk about the suggested
load balancing algorithm. Fig. 1 succinctly illustrates
the joint structure. The architecture is divided into four distinct
layers, namely the Application layer, the Control layer,
the Data layer, and the Sensing layer. Creating the user interface
for application-oriented activities falls under the
purview of the application layer. At the control plane, the
SDMN is responsible for making all decisions and controlling
how they are implemented. The control plane is the
component responsible for implementing security policies,
34
including regulated rules and packet forwarding. Routers,
switches, and bridges are examples of the types of hardware
that make up the data plane, which is often referred to as the
infrastructure plane. Devices that connect to the data plane
form the sensing plane. These devices are capable of sharing
a variety of functions and executing in response to commands
from the application layer.
Load balancing and other application-oriented flexibility
are enabled by the structure of the SDMN, as shown in Fig. 2.
The SDMN architecture enables application customization,
which allows any heterogeneous service need to be met. Due
to growing data volumes, effective load balancing is critical
for heterogeneous applications. In terms of resource consumption,
throughput, and congestion, balancing contributes to the
overall optimization of the network. In addition, load balancing
technology contributes to predictive application analysis
and provides important information about the network and
the application it supports. We will refer to the structure shown
in Fig. 2, which is an example of the technique of load balancing.
As can be seen in Fig. 1, the SDN controller has the ability
to intelligently access all network information located at a
lower level while connecting to higher network layers. The
layout of the infrastructure, the programs running on it, and
the devices connected to it are all included in the information.
Using a bird's eye view of the network, the SDN controller is
able to perform load balancing as efficiently as possible. The
execution-oriented task is supported by the application's network
architecture and knowledge of the associated devices.
Learning Environment: Designation and
Optimization of the Architecture
Machine learning models often fall into three categories: supervised
learning, unsupervised learning, or reinforcement
learning. For each category, multiple approaches are being researched
and developed to maximize the effectiveness and
accuracy of the learning. For instance, if there are problems
with load balancing in the network, machine learning specialists
will need to determine the cause of the issue, i.e., they
need to select appropriate sensors (with sufficient battery
power and the required data). This is referred to as output.
However, to generate meaningful patterns that reflect how
humans make decisions, developers working on machine
learning must include features (characteristics of the application
and sensors). For the purposes of load balancing, these
features have the potential to become applications that can be
derived through process mining, i.e., block statements, use of
input datasets, condition statements, machine instructions,
and loop operation; they can be considered application-oriented
features. Network bandwidth, sensor processing speed,
instruction per clock, and memory can thus be considered
sensor-oriented features. Topology, connection parameters,
connected devices, as well as device size are considered network-oriented
features. After the input features and output
labels are defined, the selection of the algorithm and the parameters
of that algorithm must be changed to make the results
more accurate. The learning network requires a prototype that
IEEE Instrumentation & Measurement Magazine
October 2022
Instrumentation & Measurement Magazine 25-7

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-7
