Elixir

This repository contains the implementation of Elixir, a machine learning framework for control traffic prediction in SDN systems

Overview

We provide all source codes, including Elixir implementation and experiment scripts. The provided codes are organized as follows.

• Part 1 describes the source-code organization of this repository.
• Part 2 contains the steps to configure dependencies and compilation to run Elixir.
• Part 3 gives a general guide on executing the codes.

Repository organization

The repository is organized as follows:

• Elixir/: contains the core implementation of Elixir.
• Elixir/dataset_new_avg/: contains the dataset of Elixir.
  • Elixir/dataset_new_avg/: Training dataset
    • 'fl_default_of10.csv', 'fl_default_of13.csv', 'odl_default_of10.csv', 'odl_default_of13.csv', 'onos_default_of10.csv', 'onos_default_of13.csv', 'onos_proxyarp_of13.csv', 'onos_stats1_of13.csv', 'p4_default.csv'
  • dataset_new_avg/inf_realvalue/scoring/: Scroing datset
    • 'fl_default_of10.csv', 'fl_default_of13.csv', 'odl_default_of10.csv', 'odl_default_of13.csv', 'onos_default_of10.csv', 'onos_default_of13.csv', 'onos_proxyarp_of13.csv', 'onos_stats1_of13.csv', 'p4_default.csv'
    • 'fl_default_of10.xlsx', 'fl_default_of13.xlsx', 'odl_default_of10.xlsx', 'odl_default_of13.xlsx', 'onos_default_of10.xlsx', 'onos_default_of13.xlsx', 'onos_proxyarp_of13.xlsx', 'onos_stats1_of13.xlsx', 'p4_default.xlsx'
  • dataset_new_avg/inf_realvalue/evaluation/: Evaluation datset
    • 'fl_default_of10.csv', 'fl_default_of13.csv', 'odl_default_of10.csv', 'odl_default_of13.csv', 'onos_default_of10.csv', 'onos_default_of13.csv', 'onos_proxyarp_of13.csv', 'onos_stats1_of13.csv', 'p4_default.csv'
    • 'fl_default_of10.xlsx', 'fl_default_of13.xlsx', 'odl_default_of10.xlsx', 'odl_default_of13.xlsx', 'onos_default_of10.xlsx', 'onos_default_of13.xlsx', 'onos_proxyarp_of13.xlsx', 'onos_stats1_of13.xlsx', 'p4_default.xlsx'
• Elixir/dataset_new_avg/exlixir_model_save/: pre-trained model (example of trained models of model space)

Prerquisites

• Install required dependencies and libraries with conda. Please install conda
• Execute conda $conda activate
• Import conda env $conda env create -f Elixir_env.yaml
• Activate env $conda activate Elixir_env

Execution guides

1. Training (Model Space)

$ python trainmodel.py [training_dataset csv name]

• e.g., $ python trainmodel.py onos_default_of10.csv onos_default_of13.csv
• Please check location of training dataset path: dataset_new_avg/
• Training result(trained models) path: dataset_new_avg/elixir-modelsave/

2. Inference

$ python inference.py [test_type] [training_dataset csv name]

• inference dataset path: dataset_new_avg/inf_realvalue/[test_type]/

  • require both csv file and xlsx file
  • test_type: scoring or evaluation

• scoring: $ python inference.py scoring onos_default_of10.csv onos_default_of13.csv

• evaluation: $ python inference.py evaluation onos_default_of10.csv onos_default_of13.csv

• result files(xlsx): dataset_new_avg/result/

3. Model selection

1. Naive model: $ python naiiveselction.py
   • result path (txt): dataset_new_avg/result/modelselect/naiive.txt
   • result summary path (xlsx): dataset_new_avg/result/modelselect/naiiveselectModel.xlsx
2. Freq model: $ python freq_modelselection.py
   • result path: dataset_new_avg/result/modelselect/freq.txt
   • result summary path: dataset_new_avg/result/modelselect/freqselectModel.xlsx
3. Supreme model: $ python modelselection.py
   • result path: dataset_new_avg/result/modelselect/supreme.txt
   • result summary path: dataset_new_avg/result/modelselect/naiiveselectModel.xlsx

4. Prediction Accuracy Summary

1. RMSE: $ python selectedmodelonlysummary_RMSE.py

• result: dataset_new_avg/result/modelselect

2. Freq rate: $ python freqscore_FREQ.py

• result: dataset_new_avg/result/modelselect

Contacts

• This project is from Operating systems Lab, Korea University.
• Contact: yhyoo@os.korea.ac.kr
• Contributors: Yeonho Yoo, Gyeongsik Yang, Changyong Shin, Junseok Lee, Chuck Yoo (advisor)

Citation

@ARTICLE{Meteor2023,
  author={Yoo, Yeonho and Yang, Gyeongsik and Shin, Changyong and Lee, Junseok and Yoo, Chuck},
  journal={IEEE Transactions on Services Computing}, 
  title={Machine Learning-Based Prediction Models for Control Traffic in SDN Systems}, 
  year={2023},
  volume={16},
  number={6},
  pages={4389-4403},
  keywords={Predictive models;Control systems;Training;Network topology;Throughput;Prediction algorithms;Machine learning;Machine learning;software-defined networking;control traffic;prediction model formulation;prediction robustness},
  doi={10.1109/TSC.2023.3324007}}