TUFTTE: handling Traffic Uncertainty in Failure-Tolerant Traffic Engineering

1. Overview

TUFTTE is a novel TE framework with decision-focused learning. Its objective function and constraints can be flexibly specified by the network operators. It is currently built upon the cvxpylayer, but I am also exploring other efficient tools.

2. Code Structure

Tuftte
├── algorithms
│   ├── DOTE-main
│   ├── __init__.py
│   ├── DoteSolver.py
│   ├── TEAVARSolver.py
│   ├── TESolver.py
│   └── TUFTTESolver.py
├── data
│   ├── gml_gen_topo.py
│   └── snd_gen_topo.py
├── utils
│   ├── __init__.py
│   ├── CvxpySolver.py
│   ├── GurobiSolver.py
│   ├── helper.py
│   ├── history.py
│   ├── NetworkParser.py
│   ├── NetworkTopology.py
│   ├── prediction.py
│   ├── riskMetric.py
│   └── scenario.py
├── availability.py
├── benchmark_consts.py
├── dl_experiment.py
├── environment.yml
├── main.py
├── motivation.py
└── prediction_details.py

We import DOTE from their public repository without modification.

3. Getting Started

First, install the conda environment using environment.yml.

conda create -n tuftte --file environment.yml

Second, apply a license of GUROBI solver. If you don't have a license, you can use open-source CVXPY instead by replacing all the GurobiSolver() with CvxpySolver().

Third, prepare datasets. Download abilene.txt, directed-abilene-zhang-5min-over-6months-ALL-native.tar from SNDlib and move them to data/. Generate the topology by the following commands.

cd data/
python3 snd_gen_topo.py

4. Running Experiments

It is simple to conduct the experiments. The configuration of parameters is completed in benchmark_consts.py. You just need to run main.py with different values of the argument --experiment.

python3 main.py

• ./main.py --experiment ffc_experiment for Fig. 1a.
• ./main.py --experiment teavar_experiment for Fig. 1b.
• ./main.py --experiment dl_experiment for Fig. 4.
• ./main.py --experiment noise_effect for Table II.
• ./main.py --experiment availability for Fig. 5.
• ./main.py --experiment MSE_test for Fig. 6a.
• ./main.py --experiment pos_neg for Fig. 6b.
• ./main.py --experiment watch_variation for Fig. 7.

5. Citation

Please cite our paper if our contributions benefit your research.

@INPROCEEDINGS{zhao2025,
  author={Zhao, Shiju and Zheng, Jiaqi and Chen, Guihai},
  booktitle={2025 IEEE/ACM 33rd International Symposium on Quality of Service (IWQoS)}, 
  title={Rethinking Failure-Tolerant Traffic Engineering with Demand Prediction}, 
  year={2025},
  volume={},
  number={},
  pages={1-6},
  keywords={Uncertainty;Network topology;Quality of service;Telecommunication traffic;Production;Machine learning;Predictive models;Prediction algorithms;Topology;Optimization;failure tolerance;traffic engineering;machine learning;decision-focused learning},
  doi={10.1109/IWQoS65803.2025.11143354}}