We also provide a WebRTC-based implementation for Hairpin at hairpin-webrtc.

Hairpin is built on SparkRTC ns-3 library, which provides a simulation for the real-time video communication in ns-3. It provides congestion control algorithms (GCC, NADA) and forward error correction baselines. The main functionality has been implemented in the ns3-sparkrtc repository. This repository is used to generate the parameters of Hairpin and run the experiments.

Please cite our paper if you use our simulator:

@inproceedings{nsdi2024hairpin,
  title={Hairpin: Rethinking Packet Loss Recovery in Edge-based Interactive Video Streaming},
  author={Meng, Zili and Kong, Xiao and Chen, Jing and Wang, Bo and Xu, Mingwei and Han, Rui and Liu, Honghao and Arun, Venkat and Hu, Hongxin and Wei, Xue},
  booktitle={Proc. USENIX NSDI},
  year={2024}
}

The packet can be parsed with the sparkrtc.lua in Wireshark.

Install dependency

sudo apt install build-essential libboost-all-dev

Fetch the latest ns-3 source files

./setup-env.sh

Configure before building!

For ns-3.33 and below, to use c++14 and above features, should configure like this: (More information: C++ standard configuration in build system (#352) · Issues · nsnam / ns-3-dev · GitLab)

LDFLAGS="-lboost_filesystem -lboost_system" ./waf configure --cxx-standard=-std=c++17

Run the code

To simply test if the codes can run, just run at ns-allinone-3.33/ns-3.33:

./waf --run "rtc-test --vary=1 --fecPolicy=hairpin

When the process is running, the logs can be found at ns-allinone-3.33/ns-3.33/logs/. Usually the program will finish in around 5 minutes.

Reproduce the results

We use generate_conf_* to generate configuration files, where the files are later put into *.conf files. For example, to run the hairpinone baseline in the WiFi traces, just run

python run_ns3.py --conf wifi-hairpinone.conf

run_ns3.py is a wrapper to parallelly run ns-3 experiments over different traces and in parallel.

If you want to run a single code for debugging, copy one line from the conf file and use ns-3 command line to run it.

Process the results

The process_results.py is used to process the results and generate the data to plot figures in the paper.

Generate parameters for other coefficient

cd model
python fec_w_rtx.py --func offline_data -coeff 1e+00
python merge_results.py --coeff 1e+00

Currently the coefficient only supports $10^n$.