/MixerNet

Code for our CNSM'2022 paper "A Novel Network Delay Prediction Model with Mixed Multi-layer Perceptron Architecture for Edge Computing"

Primary LanguagePython

MixerNet

Code for our CNSM'2022 paper "A Novel Network Delay Prediction Model with Mixed Multi-layer Perceptron Architecture for Edge Computing"

If you find this code useful in your research please cite

@INPROCEEDINGS{fang2022a,  
  author={Fang, Honglin and Yu, Peng and Wang, Ying and Li, Wenjing and Zhou, Fanqin and Ma, Run},  
  booktitle={2022 18th International Conference on Network and Service Management (CNSM)},   
  title={A Novel Network Delay Prediction Model with Mixed Multi-layer Perceptron Architecture for Edge Computing},   
  year={2022},  
  volume={},  
  number={},  
  pages={191-197},  
  doi={10.23919/CNSM55787.2022.9964552}
}

Setup

The main environment is (latest is also available):

  • cuda 11.3
  • torch 1.8.0
  • networkx 2.6.3
  • einops 0.4.0

Prepare dataset

NSFNET and GEANT2 datasets are publicly available here

cd dataset
# For NSFNET
wget "http://knowledgedefinednetworking.org/data/datasets_v0/nsfnet.tar.gz"
tar -xvzf nsfnet.tar.gz 
# For GEANT2
wget "http://knowledgedefinednetworking.org/data/datasets_v0/geant2.tar.gz"
tar -xvzf geant2.tar.gz

Train Model

  • Train with data process(first time)
# for NSFNET
python run.py --net nsfnetbw
# for GEANT2
python run.py --net geant2bw
  • Train
# for NSFNET
python run.py --net nsfnetbw --process False
# for GEANT2
python run.py --net geant2bw --process False

Also other model parameters and training hyper-parameters can be changed by adding argparse like --lr 3e-4 --dim 64.

Related publications: