This project reproduces the key ideas from "RouteNet: Leveraging Graph Neural Networks for network modeling and optimization in SDN",
by simulating network traffic, collecting delay and jitter metrics, and training a Graph Neural Network (GNN) to predict these metrics for SDN environments.
RouteNet: Leveraging Graph Neural Networks for network modeling and optimization in SDN
- Python 3
- PyTorch 2.4+
- networkx
- pip packages listed in
model/requirements.txt, for the routenet model - Docker (used for ONOS, Mininet, and D-ITG in containerized setup)
- NVIDIA GPU recommended (optional, speeds up training)
git clone https://github.com/jaison-joseph/CS536-project.git
cd CS536-project
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txtpython3 master.py <number of nodes> [nsfnet/geant2/germany50] <number of runs per traffic intensity value> <time per run in SECONDS> <frequency of statistics in MILLISECONDS> <min traffic intensity> <max traffic intensity>cd model
python main.py- Simulations are shorter and use fewer samples than the original RouteNet study (due to resource constraints), leading to higher prediction errors.
- Occasional Mininet instability is handled by automatic restarts.
- Overall results have higher MAE and MRE compared to the paper’s benchmarks.