This repository contains the supporting simulation, benchmarking, and plotting code for the paper A Taxonomy and Comparative Analysis of IPv4 ID Selection Correctness, Security, and Performance by Joshua J. Daymude, Antonio M. Espinoza, Sean Bergen, Benjamin Mixon–Baca, Jeffrey Knockel, and Jedidiah R. Crandall.
Before running any scripts, please build and activate the necessary conda environment by running:
conda create --file requirements.yml
conda activate IPIDSurvey
Many of our results involve computationally expensive sampling or benchmarking experiments.
If you would like to reproduce our results from scratch, follow the instructions below.
Otherwise, download our pre-computed results (available here) and extract them in this directory as results/.
You can then call the corresponding plotting scripts in collisions.py (correctness), security.py (security), and benchplot.py (performance) without having to wait for the long runtimes.
collisions.py contains our correctness analysis (Section 4.2).
Run it with python collisions.py -P <num_cores>, where -P optionally specifies additional cores to speed up the calculations depending on sampling.
Use python collisions.py --help for all options.
If results/collisions/ already exists, it will use the results therein instead of calculating them from scratch.
security.py contains our security analysis (Section 4.3).
Run it with python security.py -P <num_cores>, where -P optionally specifies additional cores to speed up the calculations depending on sampling.
Use python security.py --help for all options.
If results/security/ already exists, it will use the results therein instead of calculating them from scratch.
Our runtime benchmark (Section 4.4) is implemented in C++ and was run on a 128-core machine (dual-socket 2x 64-core AMD EPYC 7713 Zen3).
If you downloaded our pre-computed results (in this case, results/benchmark/), then you can plot the outcome with python benchplot.py.
If you are trying to run the benchmark from scratch, navigate to benchmark/ and then build with ./build.sh.
Any build errors likely will have to do with your C++ version (we require C++20), or missing boost libraries (we use boost::program_options).
A successful build will create a build/ directory containing a variety of build artifacts.
View the executable's options using ./build/benchmark --help.
Our benchmark can be replicated using ./sol_bench.sh, though you may need to adjust the parameters based on your hardware.