AI-Driven Dynamic Traffic Management in Multi-Domain SDN Networks : An intelligent framework for traffic management
Ibironke, Joshua (2025)
2025
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2025-10-31
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2025103110304
https://urn.fi/URN:NBN:fi:tuni-2025103110304
Tiivistelmä
The continuous increase in data traffic and cloud services has intensified the demand for intelligent and adaptive network management. Traditional Software-Defined Networking (SDN) controllers rely on deterministic routing algorithms such as Dijkstra and Equal-Cost Multi-Path (ECMP), which lack the ability to dynamically respond to traffic fluctuations and other inter-domain coordination challenges. As controller deployments scale across multiple administrative domains, efficient traffic management requires autonomous and data-driven control mechanisms capable of improving network performance in real time.
This thesis proposes an AI-driven traffic management framework for SDN environments, integrating Reinforcement Learning (RL) into the SDN control plane. Two algorithms, namely Q-Learning and Deep Q-Learning (DQN), were implemented and trained offline using the COST-239 network topology. The framework also combines the RYU controller, OpenFlow switches, and a reinforcement learning orchestration layer that enables dynamic path selection based on link state, delay, and overall network topology.
The study’s findings show that the DQN-based agent achieved faster convergence and more stable routing performance than the tabular Q-Learning agent. The integration of reinforcement learning within the SDN control plane successfully demonstrated dynamic path selection without compromising connectivity across the network topology. These results highlight the feasibility of using RL-based methods to enhance traffic optimization and scalability in multi-domain networks.
Despite these contributions, the implementation of this framework is subject to several limitations. The models were trained offline and deployed to a single RYU controller domain. Future research should extend this framework to include online learning, multi-agent coordination, and distributed control architectures to enable real-time intelligent traffic management in AI-driven networks.
This thesis proposes an AI-driven traffic management framework for SDN environments, integrating Reinforcement Learning (RL) into the SDN control plane. Two algorithms, namely Q-Learning and Deep Q-Learning (DQN), were implemented and trained offline using the COST-239 network topology. The framework also combines the RYU controller, OpenFlow switches, and a reinforcement learning orchestration layer that enables dynamic path selection based on link state, delay, and overall network topology.
The study’s findings show that the DQN-based agent achieved faster convergence and more stable routing performance than the tabular Q-Learning agent. The integration of reinforcement learning within the SDN control plane successfully demonstrated dynamic path selection without compromising connectivity across the network topology. These results highlight the feasibility of using RL-based methods to enhance traffic optimization and scalability in multi-domain networks.
Despite these contributions, the implementation of this framework is subject to several limitations. The models were trained offline and deployed to a single RYU controller domain. Future research should extend this framework to include online learning, multi-agent coordination, and distributed control architectures to enable real-time intelligent traffic management in AI-driven networks.