Project: RePO and RePO+ Evaluation under Simulated Attacks

Reproducing and extending RePO and RePO+ evaluation in Mininet, introducing new metrics and performance analysis.

Overview

This project reproduces the results of RePO and RePO+—robust path selection algorithms—under simulated network attack scenarios using the Mininet environment. Beyond reproduction, it addresses model shortcomings by incorporating additional performance metrics and suggesting future improvements.


Key Features

  • Reproduction of RePO and RePO+ experiments under controlled attack simulations.
  • Evaluation with extended metrics like Channel State Information (CSI) and Matthews Correlation Coefficient (MCC) to gain deeper insights into model performance.
  • Visual performance assessment through scatterplots and ROC curves.
  • Formulation of potential future directions for enhanced robustness.

Approach

  1. Experiment Setup
    • Implemented RePO and RePO+ in Mininet to simulate attack scenarios.
    • Controlled network topology to isolate the effect of attacks.
  2. Metric Extension
    • Added CSI and MCC to evaluate the classification and detection capabilities more comprehensively.
    • Compared results to original evaluation metrics.
  3. Visualization
    • Generated scatterplots for CSI and MCC distribution analysis.
    • Constructed ROC curves to visualize trade-offs between true and false positive rates.

Results & Evaluation

  • Successfully reproduced baseline results for RePO and RePO+.
  • CSI and MCC revealed performance nuances missed by original metrics.
  • ROC curves indicated RePO+ consistently outperformed RePO in most attack scenarios.
  • Scatterplot patterns highlighted areas of model instability under high attack intensity.

Future Directions

  • Explore hybrid approaches combining RePO+ with adaptive path re-selection strategies.
  • Investigate deep learning–based anomaly detection to complement path selection.
  • Extend testing to real-world network traces for validation.

Conclusion

  • RePO+ offers measurable improvements over RePO in robustness against simulated attacks.
  • Incorporating richer metrics like CSI and MCC allows for more holistic evaluation.
  • Visual analysis tools enhance interpretability of network performance.

Report

Full Report PDF