The master thesis of the graduate Tabarak Yasin Khudayr, a candidate for a master’s degree, was defended on Thursday morning on February 27, 2025, at Al-Khwarizmi Central Hall and Assistant professor Ahmed Sattar Hadi, college of AL-Khwarizmi, communication and information Engineering department, at Baghdad University, was chaired the thesis committee that consisted of assistant professor Dr. Alaa Mohammed Abd Al-Hadi, Baghdad university, computer engineering department, assistant professor Dr. Muhammed Emad Abd alsattar, Nahren university, information and engineering department and supervised by Dr. Omer Ali Adhab, Al-Khwarizmi college at Baghdad university, communication and information Engineering department . After the candidate had finished the defense, the chair of the committee announced the degree of) pass with merit (, awarded to the candidate for her research findings which are summarized:

Software-Defined Networking (SDN), used recently in data centers, has obtained significant attention because of the ability to provide flexible network management. However, SDN faces a critical challenge as a traditional network which is the congestion problem. This occurs when the network is overloaded with data traffic, leading to issues such as packet loss and increased delays. This thesis focuses on utilizing machine learning (ML) algorithms to improve congestion control in SDN. A mininet simulator was used to build a tree topology to collect a dataset using the RYU controller’s ability to monitor network statistics parameters. These parameters were used to train and test ML models and determine congestion location in switches. Four ML algorithms, called Decision Tree (DT), Random Forest (RF), K-Nearest Neighbor (KNN), and Support Vector Machine (SVM), are constructed to classify and predict congestion occurrence. The performance was discovered that RF emerges as the optimal choice for real-time congestion classification and prediction, showing a balance between classification time (32.41 sec) and accuracy (83.08%).

Conclusions

1. Mininet simulator has many features that encourage its selection as an SDN simulator, the important one is that this simulator built in Python, and can run Python scripts. RYU is selected as the SDN controller due to its ability to run Python scripts that contain the trained ML algorithms.
2. In the validation phase, grid search was useful in determining which set of values for the hyper parameters have the best results in terms of timing and accuracy to estimate the model classifier. In the testing phase, SVM has an efficient classification result but has the longest time of classification making this algorithm not ideal to classify congestion in real-time applications.
3. In the evaluation phase, ML successfully handles congestion by increasing the max rate of the queue instead of the known solution in traditional network.