This repository contains the code for generating the data, running the experiments and producing the plots of the paper "Secure Summation via Subset Sums: A New Primitive for Privacy-Preserving Distributed Machine Learning". All scripts obtain their parameters from experiment_params.py, so you only need to specify them once there.

Start by installing the packages from requirements.txt.

Simply run data_generator.py, which will produce a folder data.

• Adjust the variable GUNICORN_PATH in run_experiments.py to point to your Gunicorn installation.
• If you don't want to measure traffic but only runtime, simply run run_experiments.py. It will create a folder results with the runtimes.

• Install tcpdump: sudo apt-get install tcpdump

• Set the variable TRAFFIC_CAPTURE_BREAKS in run_experiments.py to True. This will introduce 30 s breaks between the different experiments so that we can see in the traffic logs when one experiment ended and the next one started.

• Start tcpdump and then run the experiments by running run_experiments.py. To capture the data sent by the server:

  tcpdump -e -i lo 'src localhost && dst localhost && src port 5000 && tcp' > sent.txt

  To capture the data received by the server:

  tcpdump -e -i lo 'src localhost && dst localhost && dst port 5000 && tcp' > received.txt

  You might need to change lo to the name of your loopback interface (e.g., lo0 on Mac).

• Extract the time and the packet size column of tcpdump's output:

  cut -d ' ' -f 1,13 <tcpdump file> > <output file>

• Run the cells in the Jupyter notebook traffic_log_analysis.ipynb to sum up the sizes of the packets belonging to the same experiment.

Run plots.py.