DUNE-bmv2

This repository contains the software switch implementation of DUNE. It allows running Mininet-based experiments to test inference models using Bmv2.

Dependencies and Installation

To run this project, you need to set up the P4 development environment. We rely on the scripts provided by the p4-guide repository to install all the necessary dependencies (e.g., PI, BMv2, P4C, Mininet).

Run the following commands to install the dependencies:

git clone https://github.com/jafingerhut/p4-guide.git
./p4-guide/bin/install-p4dev-v10.sh

⚠️ WARNING: There is a known issue with the behavioral-model Thrift interface (p4lang/behavioral-model#1367) that prevents the controller from connecting to the switches. Until this is fixed upstream, you must apply a patch. Please refer to the Troubleshooting guide for instructions to apply the patch.

Ensure you activate the Python virtual environment created by the installation script before executing commands or make targets:

source ~/p4-guide/p4dev-python-venv/bin/activate

Required Files

Before running the experiments, ensure you have the required ToN_IoT dataset files placed in a data/ directory at the root of the project:

• data/ToN_IoT_Test_Flow_PktCounts.csv
• data/ToN_IoT_test.pcap
• data/ton_tcp_udp.pcapng (optional depending on your exact configuration)

(You can verify the expected paths in utils/params.ini)

To generate these data files, please follow the procedure described in the DUNE data generation guide.

Note: This repository will eventually become a submodule of the main nds-group/DUNE repository.

Quick Start: Minimal Example

To run a minimal smoke test and ensure the environment is correctly set up, run:

make run-smoke-test

Reproducing DUNE Paper Results

To reproduce the results presented in Table V of the DUNE paper, execute the following target:

make run-table-5

Additional Documentation

For more detailed information regarding the project's architecture, workflows, configurations, and advanced usage, please refer to the documentation in the docs/ folder:

• Project Structure & Architecture
• Workflows & Targets
• Troubleshooting & Debugging

Citation

If you use this code or framework in your research, please kindly cite our INFOCOM 2025 paper:

@INPROCEEDINGS{11044678,
  author={Bütün, Beyza and De Andres Hernandez, David and Gucciardo, Michele and Fiore, Marco},
  booktitle={IEEE INFOCOM 2025 - IEEE Conference on Computer Communications}, 
  title={DUNE: Distributed Inference in the User Plane}, 
  year={2025},
  volume={},
  number={},
  pages={1-10},
  keywords={Sequential analysis;Accuracy;Computational modeling;Scalability;Memory management;Machine learning;Hardware;Delays;Resource management;Monitoring},
  doi={10.1109/INFOCOM55648.2025.11044678}
}