VladaSedlacek / mq-comparison-suite

A wrapper of MQ solver implementations, allowing a unified interface and performance comparison.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

MQ Comparison Suite

This project aims to provide a simple unified interface for state-of-the-art solvers for the multivariate quadratic (MQ) problem, and to facilitate comparing their practical performance. Pull requests are welcome!

Solvers

Currently, the following solver implementations are supported:

  • crossbred (the original version, not public)
  • cryptominisat (SAT solver for crypto problems)
  • libfes (fast exhaustive search)
  • magma (using the F4 algorithm for Groebner bases, closed source)
  • mq (simple deterministic algorithm)
  • mqsolver (GPU version of crossbred)
  • wdsat (SAT solver for systems coming from Weil descent)
  • XL (might need a few few tweaks to run with modern compilers and Sage 9)

The solvers need to be installed externally (though none of them are required).

Equation systems

For best results, this comparison suite should be used for equations over either $\mathbb{F}_2$ and $\mathbb{F}_{16}$. Note that Magma supports all finite fields, XL supports $\mathbb{F}_2$, $\mathbb{F}_{16}$ and $\mathbb{F}_{31}$, while the other solvers support only $\mathbb{F}_2$. However, using the Weil descent, we can turn any system of $m$ equations with $n$ variables over $\mathbb{F}_{2^d}$ into a system of $dm$ equations with $dn$ variables over $\mathbb{F}_2$. This is done by default where needed to allow comparisons.

Since some of the solvers exit after finding the first solution, this tool works for best for systems with a single solution.

While the top-level comparison is done on equation systems coming from Rainbow attacks, these can be replaced by any other systems, e.g., from the Fukuoka challenge.

Scripts

Default paths and solvers can be configured in config_utils.py. The callable scripts (via click) include:

  • rainbow_attacks.sage generates an instance of the Rainbow cryptosystem, mounts the differential attack and saves the resulting system in different formats for further usage.
    • example to just generate systems: sage rainbow_attacks.sage --seed 0 --q 2 --o2 6 --gen_only
    • example to also start solving: sage rainbow_attacks.sage --seed 0 --q 2 --o2 6 --solver cms
    • see sage rainbow_attacks.sage --help for more
  • invoke_solver.py calls a solver on a provided equation file. It is called internally by rainbow_attacks.sage, but can also be used for custom equation systems.
    • example: python3 invoke_solver.py --solver xl --q 2 --m 11 --n 4 -e my_equation_system
    • see python3 invoke_solver.py --help for usage
  • check_solution.sage checks the log output after a solver call and compares it with an expected solution. It is called internally by other scripts.
  • solve_fukuoka.sage is a simple demo to wrap solving Fukuoka challenges in a uniform way.
    • example: sage ./solve_fukuoka.sage --fukuoka_challenge_path ../fukuoka/ToyExample-type1-n10/toy_example/ToyExample-type1-n10-seed0 --solver libfes loads the Fukuoka challenge at the given path, converts it to all other formats and then tries to solve it with libfes
  • compile_solver.py takes care of solver which need to be precompiled in advance. It is called internally by other scripts.
  • compare_solvers.py is the top layer wrapper which performs a comprehensive comparison of all the solvers and logs the results.
    • example: sage compare_solvers.py --q 2 -s magma -t 3600 runs the comparison of all solvers over $\mathbb{F}_2$ except for magma, with one hour timeout for each solver

Equation file formats

Solver Extension
cb_gpu .cb_gpu
cb_orig .cb_orig
cms .cnf
libfes .mq
magma .magma
mq .mq
wdsat .anf
xl .xl

About

A wrapper of MQ solver implementations, allowing a unified interface and performance comparison.

License:MIT License


Languages

Language:Sage 57.0%Language:Python 43.0%