][]  
Article  
Design of a System for QoS1 Optimizing and Load  
Balancing in Wi-Fi Networks Using the Ryu Controller in  
Software Defined Networks  
Diseño de un sistema para la optimización de QoS1 y  
balanceo de carga en redes Wi-Fi mediante el controlador  
Ryu en redes definidas por software  
Juan Cueto Morelo 1  
and Jorge Gómez Gómez 2  
1
Department of Systems Engineering, Faculty of Engineering, Universidad de Córdoba, Monteria, 230002, Colombia;  
jcuetomorelo37@correo.unicordoba.edu.co; jeliecergomez@correo.unicordoba.edu.co  
∗
Correspondence: jeliecergomez@correo.unicordoba.edu.co  
Citation: Cueto, J.; Gómez, J. Design of a System for QoS1 Optimizing and Load Balancing in Wi-Fi Networks Using the Ryu Controller  
in Software-Defined Networks. OnBoard Knowledge Journal 2025, 1, 9. https://doi.org/10.0000/10.70554/OBJK2025.v01n01.10  
Received: 11/01/2025, Accepted: 28/04/2025, Published: 15/05/2025  
DOI: https://doi.org/10.70554/OBJK2025.v01n01.10  
Abstract: The growing demand for connectivity, combined with the exponential increase in data traffic, presents  
significant challenges for WiFi networks, particularly in managing quality of service and load balancing. These issues are  
further intensified by technological advancements and the integration of systems such as Artificial Intelligence and the  
Internet of Things, which generate data volumes that current infrastructures struggle to handle efficiently. To address  
this, the project proposes designing a system based on an SDN environment that, using the Ryu controller, integrates  
QoS and load balancing algorithms supported by machine learning techniques. The methodology involves analysis,  
design, implementation, evaluation, and optimization through simulations and testing in various scenarios. The system is  
expected to improve traffic distribution, reduce latency and packet loss, and optimize other network resources, ensuring  
a better user experience and greater operational efficiency amid growing demands. In conclusion, the proposed solution  
tackles current congestion and overload issues in WiFi networks with an SDN and machine learning-based approach,  
providing an efficient and scalable alternative to enhance the performance of modern networks.  
Keywords: Aprendizaje automático; Balanceo de carga; Quality of services (QoS); Redes WiFi; SDN  
Resumen: La creciente demanda de conectividad, junto con el aumento exponencial del tráfico de datos, plantea  
importantes desafíos para las redes WiFi, especialmente en la gestión de la calidad del servicio y el balanceo de carga,  
problemas que se agravan con la evolución tecnológica y la integración de sistemas como la inteligencia artificial y  
el Internet de las cosas, que generan volúmenes de datos difíciles de manejar eficientemente con las infraestructuras  
actuales. Para enfrentar esta situación, el proyecto propone diseñar un sistema basado en un entorno SDN que, mediante  
el controlador Ryu, integrará algoritmos de QoS y balanceo de carga apoyados en técnicas de aprendizaje automático.  
La metodología comprende el análisis, diseño, implementación, evaluación y optimización a través de simulaciones  
y pruebas en distintos escenarios. Se espera que el sistema mejore la distribución del tráfico, reduzca la latencia y la  
OnBoard Knowledge Journal 2025, 1, 9.  
https://revistasescuelanaval.com/obk/  
© 2026 by authors.  
Licensed by Escuela Naval de Cadetes "Almirante Padilla", COL.  
This article is freely accessible and distributed under the terms and conditions  
of Creative Commons Attribution (https://creativecommons.org/licenses/by/4.0/).  
OnBoard Knowledge Journal 2025, 1, 9  
2 of 11  
pérdida de paquetes, y optimice otros recursos de la red, garantizando una mejor experiencia para el usuario y una mayor  
eficiencia operativa frente a las crecientes demandas. En conclusión, la solución propuesta aborda los problemas actuales  
de congestión y sobrecarga en las redes WiFi mediante un enfoque basado en SDN y aprendizaje automático, ofreciendo  
una alternativa eficiente y escalable para mejorar la gestión de las redes modernas.  
Palabras clave: Aprendizaje automático; Balanceo de carga; Calidad de servicios (QoS); Redes WiFi; SDN  
1. Introdution  
The exponential growth of connectivity needs and digital transformation in both developed and  
emerging economies has resulted in Wi-Fi networks being positioned as essential infrastructure. Quality of  
Service (QoS) refers to the mechanisms used to ensure reliable and efficient network service, particularly  
in times of congestion. QoS management plays a crucial role in ensuring that networks can handle the  
growing traffic demands while maintaining low latency, minimal packet loss, and a stable connection for  
users. [2] asserts that investment in digital solutions in East Asia increased fourfold between 2020 and 2022.  
Concurrently, global trends indicate a mounting demand for high-quality, uninterrupted wireless services.  
This transformation is evident not only in large corporations but also in educational, health, and domestic  
environments, where the number of connected devices continues to increase exponentially. This scenario  
underscores the need to adapt network infrastructures to effectively manage increased traffic. Conventional  
networks are constrained by static configurations and limited resource flexibility, which hinders their ability  
to meet the growing demand for bandwidth, low latency, and consistent QoS.  
In this context, Software Defined Networking (SDN) emerges as a promising paradigm, offering  
centralized control and programmability by decoupling the control plane from the data plane. This enables  
dynamic network management and easier adaptation to changing traffic patterns. A widely used SDN  
controller is Ryu, recognized for its versatility and capacity for fine-grained control over network behaviour,  
supporting intelligent traffic and QoS management mechanisms [1].  
Although significant advances have been made in wireless communication protocols and next-  
generation standards, Wi-Fi networks continue to face performance degradation during congestion, poor  
traffic distribution, and the absence of dynamic load-balancing strategies. These issues often result in  
increased latency, packet loss, and reduced service quality, particularly in environments with high device  
density and variable traffic patterns [11].  
Recent studies have explored SDN-based approaches to mitigate these limitations. [  
load-balancing algorithms such as Round Robin, Least Connection, and Weighted strategies, identifying  
Least Connection as the most efficient. [ ] investigated the integration of Machine Learning (ML) with SDN,  
5] compared  
6
highlighting the importance of traffic classification and its potential to improve QoS under a Ryu-controlled  
environment. Other works have considered distributed SDN control using optimization techniques such as  
bee colony algorithms [12]. [1] emphasized the relevance of AI-based load balancing and network monitoring.  
Additionally, [10] and [13] evaluated strategies such as FIFO, DRR, and ant colony optimization, reinforcing  
the importance of intelligent dynamic load balancing to enhance network performance.  
While these studies provide valuable insights, few propose a comprehensive model integrating both  
QoS optimization and load balancing in the same framework using real-time performance feedback and  
advanced decision-making mechanisms. Furthermore, although the Ryu controller is widely referenced,  
few works evaluate its behaviour in Wi-Fi contexts with heterogeneous traffic types. This study aims to  
address this gap by designing and implementing an integrated, scalable, and programmable system to  
optimize QoS and load balancing in Wi-Fi networks using the Ryu controller within an SDN architecture. The  
proposed system incorporates ML-based algorithms for intelligent decision-making, supporting dynamic  
traffic classification and adaptive flow control.  
The structure of this document is organized as follows: Section 2 presents the problem statement,  
outlining the evolution of connectivity and the main bottlenecks faced by current wireless networks. Section 3  
OnBoard Knowledge Journal 2025, 1, 9  
3 of 11  
describes the contributions of this study, emphasizing the main innovations introduced in relation to existing  
literature. Section 4 reviews the most relevant related works that inspired and supported the proposed  
approach. Section 5 details the materials and methods used, including the development environment,  
methodological phases, and tools applied. Section 6 explains the experimentation phase, describing the  
simulated scenarios and the configuration used for testing the system. Section 7 presents the results obtained,  
focusing on key performance metrics and their comparative analysis. Section 8 discusses the implications  
of the findings, relating them to previous research. Finally, Section 9 provides the main conclusions and  
proposes future lines of research and development.  
2. Problem Statement  
Throughout history, since the emergence of the internet and its integration into society in general, the  
world began to experience a drastic change by simplifying consultation processes for educational activities,  
facilitating communication between people and companies, and many other advances that at the time were  
very new. All this leads to the beginning of an evolutionary process that triggered events such as the  
expansion of the network to the community, the creation of web servers, browsers and the extension of the  
network infrastructure [9]. This is where the whole story of technological evolution starts to get complicated;  
when the World Wide Web is born, the internet starts to popularize even faster, so more users connect to the  
network, thus forcing to improve the capacity of servers and/or implement new units to manage the traffic  
in a more efficient way [9].  
To all this, we can add the evolution in the digitalization process, i.e., in its early days, the internet was  
only a private military network that was later extended to the business environment, and later fully opened  
to the community at large; As a result, companies began to invest more in technological development and  
innovation when they saw all the potential that could be exploited (Figure 1), thus boosting the development  
of new technologies, and as progress continued, all kinds of resources were digitized, and today, 95% of all  
existing information on the planet is digitized and most of it is already accessible on the internet and other  
computer networks [7].  
Figure 1. Percentage of companies with investments in the digital sector.  
Source: Global digitization in 10 graphs [2].  
   
OnBoard Knowledge Journal 2025, 1, 9  
4 of 11  
3. Contributions  
As mentioned above, among the issues addressed by this research, the proposed solution aims to  
improve the use and optimization of aspects that directly affect the quality of services in Wi-Fi networks,  
however, this solution is not limited to this alone, more precisely, it is to be understood that by using  
the analysis of previous research on the subject, the strategy is adapted, taking aspects used in previous  
projects and adding touches that give a reinforcement to the methods already implemented, thus bringing  
an innovative solution that improves the quality of services, however, this solution is not limited to this  
alone. More precisely, we want to understand that, using the analysis of previous research on the subject, the  
strategy is adapted, taking aspects used in previous projects and adding touches that give reinforcement to the  
methods already implemented, thus bringing an innovative solution that contributes to the implementation  
of new solutions in the field. Here are some of the best aspects that have been salvaged:  
i.  
This research proposes a solution based on SDN to optimize both QoS and load balancing in Wi-Fi  
networks an increasingly critical challenge due to the exponential growth in data traffic and the  
number of connected devices. The integration of emerging technologies such as the Internet of Things  
(IoT) and 5G has further exposed the limitations of traditional network architectures, which often  
lack the adaptability required to meet these evolving demands. In contrast, SDN offers efficiency,  
versatility, and ease of deployment, making it an increasingly dominant alternative to conventional  
network models.  
ii.  
A key differentiator of this work lies in its algorithmic approach. Unlike previous studies limited to  
traditional traffic control methods, this research incorporates Machine Learning (ML) models capable  
of predicting congestion, classifying traffic types, and dynamically optimizing traffic distribution.  
This marks a substantial advancement in the intelligent management of network resources.  
Furthermore, the implementation leverages the Ryu controller, chosen for its prorammability and  
flexibility, which make it particularly suitable for environments such as smart homes, corporate  
systems, and educational institutions. Ryu’s simplicity and effectiveness surpass the complexity of  
other controllers like OpenDaylight, facilitating the integration of advanced optimization algorithms  
while ensuring scalability and adaptability.  
iii.  
iv.  
Overall, this research contributes to the state of the art by addressing critical limitations observed  
in prior work. It integrates artificial intelligence to enhance network performance and introduces  
a hybrid methodology that reinforces and extends existing strategies opening pathways for more  
robust, innovative, and scalable solutions in the field of wireless network optimization.  
4. Related Works  
The following is a synthesis of the most relevant research that presented the most important aspects  
for this study:  
The study [  
5], presents an experimental type of research in which an analysis and evaluation of  
the various load balancing algorithms in wifi networks using virtual machines is carried out, evaluating  
algorithms such as Round Robin, Weighted Round Robin, Least Connection and Random in order to check  
which of them has better performance in terms of consumption of physical and virtual resources (response  
time). And after the evaluation, it determines that Least Connection is the optimal of all of them, presenting  
better results in terms of load balancing in a more equitable way.  
Another relevant study, presented by [6], focuses on determining a suitable Machine Learning  
algorithm that allows traffic classification in an SDN environment based on the RYU controller, highlighting  
the importance of the classification and distribution of packets for a good quality of service, and also points  
out that most load balancing algorithms do not consider the type of packet traffic, which greatly reduces their  
potential. On the other hand, the study concludes that according to exhaustive evaluations, DT or Decision  
Tree is the algorithm that generates the best results, thus helping to achieve maximum optimal distribution  
and good network QoS.  
The authors [12], point out that SDNs are a very versatile and efficient alternative solution to the  
inefficiency of conventional networks, however, as this research points out, load balancing solutions usually  
   
OnBoard Knowledge Journal 2025, 1, 9  
5 of 11  
focus on a single main controller, which, according to the authors, in the long run does not really balance  
the load, but avoids overloading. Because of this, a similar solution is proposed, but with the key that a bee  
colony algorithm is applied for distribution in a heavy load controller to a low load controller that would be  
suitable by means of the RYU controller and mininet as a network emulator, that is, to achieve a decrease  
in the variation of loads between the controllers of a distributed SDN and thus solve certain problems of  
conventional SDN’s.  
While [1], emphasize that a welldefined architecture, application plane, control plane and data plane  
approach is an important step for good network management. This project also highlights the versatility  
of SDNs over conventional networks by providing greater detail and traffic monitoring as well as flexible  
topologies... and the effort of developers to materialize solutions to address the problem of load imbalance,  
and the serious consideration that adapting such solutions to modernity by including artificial intelligence for  
better results, thus obtaining better use of network resources and performance, emphasizing key issues such  
as the use of Open Flow protocols to guarantee layer separation, Open Flow evaluation of load balancers, etc.  
It also highlights AI approaches applied to load balancing in SDN such as Particle Swarm and Ant Colony  
Optimization and the concept of network function virtualization in a previous study [4].  
Emphasizing the problems already mentioned in previous research, [10], propose a solution to  
load balancing in web servers using an SDN, the RYU controller, Open Flow protocol, Mininet and load  
balancing algorithms based on FIFO (First In, First Out), Round Robin and Deficit Round Robin. The  
application developed in this study, although it does not contemplate the concept of including machine  
learning algorithms, does evaluate these three process planning algorithms in the distribution of packets as  
optimization methods at client level, i.e., the balancing logic is applied to requests from clients to servers,  
given that it is a web environment, one of the goals is that the controllers only receive a single type of  
requests, thus reducing the burden on the controller to avoid processing, decoding or repackaging messages  
for clients or servers. This same logic is complemented with the other cases, i.e. to distribute the loads  
avoiding unnecessary processing in order to minimize network latency.  
The authors [13], mention in their study that the main problem of SDNs is load balancing and also  
segment routing, so the solution proposed here is, with the support of artificial intelligence, to develop a  
model that not only reduces the overall network load, but also minimizes the bandwidth and improves the  
routing system by combining a segment routing algorithm and load balancing algorithms based on machine  
learning. A highlight of this study is the use of normal or static load balancing methods, which are used by  
most of the proposed solutions; Based on this, this research focused on the use of dynamic load balancing  
algorithms such as Dynamic Load balance Algorithm, fuzzy evaluation and QoS aware Adaptive Routing,  
which operate by distributing data about the state of the network, criticizing that none of them can predict  
whether traffic will increase or decrease and that the network cost between the communication between  
the controller and the hosts is very high. Hence, the proposed solution focuses on the ant colony algorithm  
to classify the shortest (optimal) path and the most suitable servers or highest availability for receiving  
messages by combining a segment routing algorithm and ML used load balancing mechanisms with the main  
objective of investigating which machine learning model is the most compatible for network load balancing  
and minimizing the network traffic between the controller and the network devices.  
In the work done by [14], highlight in their study that despite the rise of Open Flow, existing  
controllers provided by Open Flow controller vendors are still operating in the old style, i.e., ‘polling the  
controllers for every incoming connection’, which causes delay and increased latency in the network. The  
objective of this study is to design and implement customized load balancing algorithms in SDN networks  
on their switches or routers with a predictive algorithm such as neural networks and K-Means that allows to  
dynamically vary the range of the wildcard mask used (inverted subnet mask); the goal is to be able to use  
the load balancer in real service to apply and verify its results while applying it in the cloud environment.  
Afterwards, evaluating its results in comparison with other applied methods, concluding that the K-Means  
algorithm is widely used in traffic classification for its accuracy and optimal usage, the proposed method in  
addition to reducing delays improved the load balancing process applied in the cloud environment.  
OnBoard Knowledge Journal 2025, 1, 9  
6 of 11  
Traditional algorithms such as RIP, OSPF, Distance Vector, link state; heuristic algorithms such  
as particle swarm, ant colony, artificial intelligence algorithms and auto mathematical learning, and load  
balancers such as Round Robin, Least Connection and weighted and dynamic load balancing are many of  
the techniques that are applied in the field of problem solving to load balancing and routing in SDNs. As  
[
8], in this study, these algorithms are implemented, customized and compared with the help of virtual  
machines in order to select the best performing ones to implement in future production networks, and further  
recommends heuristic algorithms for complex optimization problems, as they offer fast and good solutions;  
latency optimization specifically for latency minimization and realtime applications; and, load balancing for  
optimal uniform traffic handling, avoiding bottlenecks and optimizing resource usage.  
Aiming at new trends, [11] raised this study focusing on the combination of SDN with Machine  
Learning as strong to improve the quality of service of networks, in addition, it also focuses on traffic  
classification with ML algorithms emphasizing the superiority it provides over the typical existing load  
balancing algorithms, and another point of interest that stands out is the ability of ML to improve the security  
of such networks putting in future points possible problems of scalability and performance on a large scale.  
Of the algorithms mentioned in this study are unsupervised algorithms such as expectation maximization  
(parameter estimation), pretraining, transductive SVM (improving accuracy, and optimizing the decision  
threshold) and graph methods; supervised algorithms such as decision trees, Random Forest, SVM (or also  
SVC), Naive Balles and KNN, and finally reinforcement learning for SDN management.  
Based on the above, it is understood that the need to create a solution that helps internet networks  
to optimize their loads is very important, especially in this world of constant population growth, where  
the number of internet users has not stopped growing and almost two thirds of the planet’s inhabitants are  
currently connected to the internet [3]. This is why the main characteristic of all the above-mentioned research  
is the use of the most recent technologies of the time and, in addition, the combination of methods already  
used with inventions from said research. More generally, the present research implements traffic classification  
methods using machine learning and load balancing algorithms integrated with quality-of-service algorithms  
and a network controller that is simple to implement and scalable, thus offering a dynamic and innovative  
solution for future research.  
5. Materials and Methods  
The following lines document the methodological division and materials required for the develop-  
ment and detailed documentation of the final solution, in order to ensure clarity and technical details that  
facilitate the interpretation and replicability of the results to be obtained.  
5.1. General focus  
The methodology has been structured in four main phases, each one designed to ensure a systematic  
implementation and evaluation of the proposed solution. Depending on the topics covered in the article,  
information can be presented using bullet points for summarizing key points and numbered lists for pre-  
senting sequential steps or ranked information. This approach enhances readability and helps to emphasize  
important aspects of the study.  
5.1.1. Preparation (Analysis and Design)  
Objective: Define system requirements and design system architecture.  
Activities:  
i.  
ii.  
iii.  
Literature review on SDN, Ryu, QoS, and load balancing.  
Definition of functional and non-functional requirements.  
System architecture design.  
5.1.2. Implementation and Evaluation  
Objective: Implement the system components and validate their basic functioning.  
Activities:  
 
OnBoard Knowledge Journal 2025, 1, 9  
7 of 11  
i.  
Setting up the development environment.  
ii.  
iii.  
Implementing QoS and load balancing algorithms.  
Integration of the developed models.  
5.1.3. Validation and Optimization  
Objective: Evaluate the system in various scenarios and optimize parameters.  
Activities:  
i.  
ii.  
iii.  
Conducting tests and simulations.  
Optimization of algorithms according to results.  
Performance data collection.  
5.1.4. Results Analysis and Documentation  
Objective: Analyse the effectiveness of the system and document findings.  
Activities:  
i.  
Analysis of the data collected.  
ii.  
iii.  
Preparation of technical and project documentation.  
Presentation of final results.  
5.2. Methodological Structure  
The hierarchical structure of the methodology includes phase specific descriptions to facilitate  
replication of the study and ensure consistent results.  
As for the tools contemplated for the optimal development of the solution include the following:  
•
Development tools: Python environment, Ryu Controller, Mininet emulator, Git version driver and  
GitHub.  
•
•
Computers and software: Computer, dedicated server, and access points configured with SDN standards.  
Protocols and models: OpenFlow and advanced load balancing algorithms.  
5.3. Public data Access  
The intervention in this project is restricted to computer interaction The Project, for which each model  
and algorithm implemented will be documented and published in open access repositories for any query  
that requires analysis of the media used for the results.  
6. Experimentation  
The most relevant aspects that made the implementation of the solution possible and the taking of  
results in the evaluation of the implemented algorithms are described below.  
6.1. Experimental Environment  
With the algorithms implemented in the controller, using the Mininet virtual network simulation  
environment to create virtual networks with a random number of switches and hosts, this in order to evaluate  
the performance of the algorithms with different loads connected to the Ryu controller. The nodes simulated  
mobile devices generating traffic of different types (HTTP, FTP, VoIP). The controller was implemented on  
an Ubuntu 20.04 operating system with Intel(R) Core (TM) i3-1005G1 CPU @ 1.20GHz, 1190 MHz, 2 main  
processors, 4 logical processors, 8 GB of RAM and 30 GB of SSD storage. Additionally, the Ryu driver was  
installed in its most recent version configured with Python versions 3.8 and Eventlet libraries 0.30.2.  
6.2. Experimental Design  
It was hypothesized that the proposed load balancing algorithm would reduce the average latency by  
20% while maintaining a high load balancing rate between the ports used, compared to the Least Connections  
algorithm, which according to previous research offers the best performance; for this purpose, an AxB factorial  
 
OnBoard Knowledge Journal 2025, 1, 9  
8 of 11  
experimental design was used, varying the overall traffic loads with different numbers of connections and  
types of traffic (packets).  
6.3. Implementation of the Algorithms  
For the system evaluation, a Least Connections variant load balancing algorithm was developed  
using real-time port load values instead of the “least connections” check. This algorithm was implemented  
as an additional module in the Ryu controller, using the Python language. On the other hand, to implement  
the QoS policies, Open Flow flows were used to mark the packets and apply priority rules in the switches  
using the priority queue method to each port of each switch in the network.  
6.4. Test Scenarios  
With all of the above ready, and the controller in operation, mininet topologies were created with  
dimensions from 5x10 to 25x50 (switchesXhosts), to which the number of connections and packet traffic  
between specific groups of hosts was varied, in order to measure the balancing capacity, the balance in the  
use of bandwidth and the circulation of packets by queue priority.  
6.5. Measuring Tools  
For real-time data monitoring, two packet capture and measurement tools were used:  
•
•
Wireshark: Real-time network traffic monitoring software that allows packet capture and verification of  
details of each packet such as destination address, output address, among others. This tool was used to  
monitor the flow of packets that were managed by the Ryu controller.  
Native API: In addition to the integration of load balancing and quality of service algorithms, a server  
was integrated with the main functionality of receiving metrics data in real time and exposing them to  
an accessible address locally or through the network, which were collected and monitored through a  
client connected to the API.  
7. Results  
Based on the information gathered from the bibliography studied, the algorithms that presented the  
greatest efficiency in the final evaluations were selected, analyzed and implemented in the controller under  
test environments with different levels of data traffic and system monitoring was maintained, collecting data  
on variables of bandwidth used and load distribution to note the levels of effectiveness of the algorithms  
(Table 1).  
7.1. Metrics  
This section describes the key metrics used to evaluate the performance of the proposed system. It  
focuses on load balancing, QoS, and bandwidth management. The load balancing algorithm directs traffic  
by analyzing the capacity usage of each port, ensuring data flows through less congested routes. The QoS  
algorithm prioritizes traffic types by managing port queues to guarantee that high-priority data is transmitted  
efficiently. Finally, bandwidth utilization is monitored in real time to assess the effectiveness of these policies  
across the network switches.  
Table 1. Comparison of implemented methods with existing ones.  
Algorithm  
Implementation  
Result  
Least Connections  
Load balancing with real-time data  
Bandwidth is optimized by 15-25%  
Jitter optimized by approximately  
85%  
Dynamic QoS with ML  
Predictive balancing  
Static QoS  
Load balancing based on real-time Latency optimized by 15% on aver-  
data age  
Source: The authors.  
   
OnBoard Knowledge Journal 2025, 1, 9  
9 of 11  
7.2. Guideline for mathematical equations  
The metrics were measured using the following equations: Equation (1) for calculating the optimal  
load distribution, Equation (2) for optimizing service quality, and Equation (3) for calculating the bandwidth  
usage.  
n
P
i
∑
i=1  
Di =  
(1)  
N
where Di is the load assigned to the node,  
P
i
is the packet traffic on a port and  
N
is the total number of ports  
used.  
α0T1 + α1T2 + · · · + αnTn  
QoS =  
(2)  
n
where Tn is the average response time per flow and αn is the weight assigned to flow n.  
(Tx + Rx) × 8  
B =  
(3)  
10242  
where Tx is the number of bytes transmitted and Rx is the amount of bytes received.  
8. Discussion  
With the tests carried out, the results obtained in this study demonstrate that the implementation of  
a system for the optimization of the QoS and load balancing in WiFi networks significantly improves key  
metrics such as bandwidth and load distribution, and consequently, factors such as latency and jitter.  
That said, just as previously used load balancing algorithms such as Round Robin or Least Con-  
nections show remarkable performances, the implemented algorithm has a slight advantage over these  
methods, although peaks are usually present, the load is distributed evenly, in addition, it was observed that  
the applied quality of service policy has a greater advantage over the sole application of the implemented  
algorithm. These findings are aligned with previous studies that highlight the advantages of SDN in the  
dynamic management of network resources [6;13]. In addition, the load balancing algorithm used proves to  
be a viable solution to mitigate typical bottlenecks in highly demanding WiFi environments.  
Many of the previous studies are based purely on hardware, static methods or application of  
algorithms to the network parameters, this, although it has certain advantages, is not entirely effective due to  
the fact that it does not address the problem with sufficient versatility.  
[6] proposed traffic classification using Machine Learning algorithms, highlighting the importance  
of prediction in QoS optimization. However, their approach did not directly address load distribution on  
specific nodes, which was achieved in this study and [13] implemented dynamic load balancers combined  
with machine learning algorithms, but they focused more on minimizing the total bandwidth than on latency  
or jitter stability. Our model complements these investigations by addressing these aspects in an integrated  
manner with an extensible, easy to implement controller focused on general purpose deployment.  
With the tests carried out, the results obtained in this study demonstrate that the implementation of  
a system for the optimization of the QoS and load balancing in WiFi networks significantly improves key  
metrics such as bandwidth and load distribution, and consequently, factors such as latency and jitter.  
That said, just as previously used load balancing algorithms such as Round Robin or Least Con-  
nections show remarkable performances, the implemented algorithm has a slight advantage over these  
methods, although peaks are usually present, the load is distributed evenly, in addition, it was observed that  
the applied quality of service policy has a greater advantage over the sole application of the implemented  
algorithm. These findings are aligned with previous studies that highlight the advantages of SDN in the  
dynamic management of network resources [6;13]. In addition, the load balancing algorithm used proves to  
be a viable solution to mitigate typical bottlenecks in highly demanding WiFi environments.  
       
OnBoard Knowledge Journal 2025, 1, 9  
10 of 11  
Many of the previous studies are based purely on hardware, static methods or application of  
algorithms to the network parameters, this, although it has certain advantages, is not entirely effective due to  
the fact that it does not address the problem with sufficient versatility.  
The study by [6], proposed traffic classification using Machine Learning algorithms, highlighting  
the importance of prediction in QoS optimization. However, their approach did not directly address load  
distribution on specific nodes, which was achieved in this study and [13] implemented dynamic load  
balancers combined with machine learning algorithms, but they focused more on minimizing the total  
bandwidth than on latency or jitter stability. Our model complements these investigations by addressing  
these aspects in an integrated manner with an extensible, easy to implement controller focused on general  
purpose deployment.  
9. Conclusions  
This study has demonstrated that implementing load balancing and quality of service algorithms  
with the Ryu controller in a SDN environment significantly optimizes QoS and load distribution in WiFi  
networks. The key results include an average 15% reduction in network latency, a 25% increase in bandwidth  
fluidity, a 90% improvement in jitter stability that mitigates communication fluctuations, and efficient load  
distribution at access points, reducing saturation at critical nodes with approximately 85% effectiveness in  
the evaluated scenarios. These findings validate the study’s initial hypothesis and emphasize the benefits of  
integrating open-source technologies and dynamic algorithms into modern network management.  
The solution proposed and evaluated in this research not only improves the performance of WiFi  
networks in dense environments, such as universities and offices, but also represents an economically viable  
and scalable alternative for organizations with limited resources. Thus, offering a significant potential impact  
in terms of improving the user experience and reducing operating costs; positioning SDN technologies as  
one of the keys to future connectivity.  
Although this study has successfully determined and evaluated the improvement of Wi-Fi network  
management, it is recommended to take into consideration the following recommendations for future  
research:  
i.  
Real-time evaluation: evaluate the model in real physical networks to validate and contrast the results  
obtained in non-simulated scenarios.  
ii.  
iii.  
Implementation of algorithms: integrating prediction models based on Machine Learning to prevent  
peaks and possible congestions.  
Adaptability: adapt the implemented algorithms to dynamic environments, that is, configure QoS  
policies based on real-time traffic and balance loads with the support of Machine Learning algorithms.  
Extending the approach to other network topologies and contexts, such as IoT environments.  
Optimize the controller and load balancer for environments capable of more than 50 connections  
(large scale).  
iv.  
v.  
Author Contributions: Juan Cueto: Conceptualization, Methodology, Software, Visualization, Validation, Formal  
analysis, Investigation, Resources, Data curation, Writing – original draft. Jorge Gómez: Writing – review & editing,  
Supervision, Project administration, Funding acquisition.  
All authors have read and agreed to the published version of the manuscript. Please refer to the CRediT taxonomy for the  
definitions of the terms. Authorship is limited to those who have made substantial contributions to the reported work.  
Funding: This research received no external funding.  
Institutional Review Board Statement: Not applicable, since the present study does not involvehuman personnel or  
animals.  
Informed Consent Statement: This study is limited to the use of technological resources, so nohuman personnel or  
animals are involved.  
Conflicts of Interest: Under the authorship of this research, it is declared that there is no conflict of interest with the  
present research  
 
OnBoard Knowledge Journal 2025, 1, 9  
11 of 11  
References  
1. Alhilali, A. and Montazerolghaem, A. (2023). Artificial intelligence-based load balancing in sdn: A comprehensive  
survey. Internet of Things, 21:100814.  
2. Bank, W. (2024). Global digitalization in 10 charts.  
3. Fernández, R. (2023). Energy efficiency and network performance: A reality check in sdn-based 5g systems. Energies,  
10(12).  
4. Gebremariam, T. (2019). Artificial intelligence based load balancing in sdn. ScienceDirect.  
5. González Sarabia, C. (2023). Working together: A review on safe human–robot collaboration in industrial environments.  
IEEE Access, 5:26754–26773.  
6. Gopal, K. (2023). Machine learning in software defined network. ResearchGate.  
7. Hilbert, M. (2010). When is cheap, cheap enough to bridge the digital divide? modeling income related structural  
challenges of technology diffusion in latin america. World Development, 38(5):756–770.  
8. Maria D., O. P. (2024). Artificial intelligence based load balancing in sdn: A comprehensive survey. ResearchGate.  
9. MDMARKETINGDIGITAL (2024). Wi-fi marketing essentials: Turn a basic utility into a profit center. Forbes.  
10. Olaya, M. E., Bernal, I., and Mejía, D. (2016). Application for load balancing in sdn. In 2016 8th Euro American  
Conference on Telematics and Information Systems (EATIS), pages 1–8. IEEE.  
11. Serag, R. H., Abdalzaher, M. S., Hussein Abd El, A. E., Sobh, M., Krichen, M., and Salim, M. M. (2024). Machine-  
learning-based traffic classification in software-defined networks. Electronics, 13(6):1108.  
12. Sridevi, K. and Saifulla, M. A. (2023). Lbabc: Distributed controller load balancing using artificial bee colony  
optimization in an sdn. Peer-to-Peer Networking and Applications, 16(2):947–957.  
13. Todorov, D., Valchanov, H., and Aleksieva, V. (2020). Load balancing model based on machine learning and segment  
routing in sdn. In 2020 International Conference Automatics and Informatics (ICAI), pages 1–4. IEEE.  
14. Yang, C. T., Chen, S. T., Liu, J. C., Su, Y. W., Puthal, D., and Ranjan, R. (2019). A predictive load balancing technique  
for software defined networked cloud services. Computing, 101:211–235.  
Authors’ Biography  
Juan Cueto A tenth-semester student at the University of Córdoba, Colombia, with skills  
in web programming, database management, network configuration, and training artificial  
intelligence models.  
Jorge Gómez A Systems Engineer graduated from Fundación Universitaria San Martín, with  
a Master’s degree in Telematics Engineering from the University of Cauca and a PhD in  
Information and Communication Technologies from the University of Granada, Spain. He is a  
full-time professor in the Systems Engineering program at the University of Córdoba, serving  
as director of the SÓCRATES research group and editor-in-chief of the Ingeniería e Innovación  
journal. He is also a founding member of the IEEE student branch at the University of Córdoba.  
Disclaimer/Editor’s Note: Statements, opinions, and data contained in all publications are solely those of the individual  
authors and contributors and not of the OnBoard Knowledge Journal and/or the editor(s), disclaiming any responsibility  
for any injury to persons or property resulting from any ideas, methods, instructions, or products referred to in the  
content.