An SDN-based traffic load prediction using machine learning

Abstract

This thesis is an experimental attempt at predicting network load using machine learning techniques in an SDN (Software Defined Networking) environment. Software-defined networking (SDN) has revolutionized computer network environments by separating the control and data planes, enabling centralized control through software-defined controllers. SDN offers numerous benefits, including enhanced network configurability, simplified management, and flexibility in resource allocation based on real-time demands. However, specific challenges, such as security, reliability, and congestion, affect the efficiency and effectiveness of SDN systems. Comparison of Congestion Control: SDN vs. Traditional Networks A fundamental disparity between Software-Defined Networking (SDN) and traditional networks lies in their approaches to congestion control. Traditional networks often adopt fixed routing strategies, which can result in sub-optimal resource utilization during congestion. In contrast, SDN’s separation of control and data planes enables real-time congestion prediction and intelligent rerouting. This dynamic adaptation optimizes resource allocation, ensuring uninterrupted service during traffic peaks. The stark divergence in adaptability becomes evident when assessing how each responds to congestion events. Traditional networks grapple with sluggish responses due to their reliance on predetermined routes. In contrast, SDN’s machine learning-driven congestion management, as showcased in this research, promptly mitigates congestion by dynamically rerouting traffic, enhancing network performance under duress. Congestion, in particular, significantly impacts the quality of SDN services. Although existing congestion avoidance schemes exist, there is room for improvement to provide high-quality service. This research proposes a congestion avoidance model that combines deep neural networks and the Mijumbi (Mijumbi et al., 2014) algorithm to address the congestion issue in SDNs. The proposed model utilizes a deep neural network algorithm to predict congestion based on an in-depth analysis of data communication within the network. The model effectively mitigates congestion and ensures smooth transmission by intelligently rerouting data through alternative routes. The subsequent stage of the model employs the Mijumbi algorithm to divert predicted congestion flows away from congested routes, thereby enabling congestion-free communication. Integrating the proposed model in SDNs results in accurate congestion prediction and efficient congestion management. The model intelligently detects and redirects data communication through alternative routes, utilizing available resources and improving network quality of service. An online dataset, specifically the DDOS dataset, evaluates the proposed model’s performance, considering parameters such as sender nodes, receiver nodes, packet size, transmission time, and protocol. The proposed model is simulated in Python Jupyter Notebook, and the results are compared with a baseline model (QRTP) to assess performance. The proposed model significantly improves quality-of-service and some metrics; including packet loss rates, throughput, reconfiguration effects, and link utilization. It also demonstrates robustness in dynamic network situations, maintaining its effectiveness despite congestion, network topology changes, and data communication route alterations. In conclusion, this research presents a novel congestion avoidance model for SDNs that integrates machine learning and deep learning techniques. The proposed model proactively addresses congestion problems, optimizes solutions, and enhances user experience and network performance. The findings contribute to developing a more reliable and scalable communication network, providing a comprehensive solution for network congestion challenges in SDNs.

Date	29 Sept 2023
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Michel Kadoch (Supervisor)