ABY 3 and Applications

Introduction

This library provides the semi-honest implementation of ABY 3 and Fast Database Joins for Secret Shared Data.

The repo includes the following application:

• Linear Regression (training/inference)
• Logistic Regression (training/inference)
• Database Inner, Left and Full Joins
• Database Union
• Set Cardinality
• Threat Log Comparison (see Section 5)
• ERIC Application (see Section 5)

A tutorial can be found here. It includes a description of how to use the API and a discussion at the end on how the framework is implemented.

Warning

This codebase should NOT be considered fully secure. It has not had a security review and there are still several security related issues that have not been fully implemented. Only use this codebase as a proof-of-concept or to benchmark the perfromance. Future work is required for this implementation to be considered secure.

Moreover, some features have not been fully developed and contains bugs. For example, the task scheduler sometime fails. This is a known issue. Consider using MP-SPDZ if you want a framework that will be well supported.

Build

The library is cross platform and has been tested on Windows and Linux. The dependencies are:

• libOTe
• Boost (networking)
• function2
• Eigen

In short, this will build the project

git clone https://github.com/ladnir/aby3.git
cd aby3/
python3 build.py --setup
python3 build.py 

To see all the command line options, execute the program

out/build/linux/frontend

or

out/build/x64-Release/frontend/frontend

The library can be linked by linking the binraries in lib/ and thirdparty/win or thirdparty/unix depending on the platform.

Additional Notes (C++20 Compatibility & Build Fix)

When building on certain environments (e.g., newer compilers or macOS), the default C++ standard setting in libOTe may cause compatibility issues with the macoro library.

Step 1: Initial Setup

python3 build.py --setup

Step 2: Fix C++ Standard Version

After completing Step 1, modify the following file:

thirdparty/libOTe/CMakeLists.txt

Change

set(LIBOTE_STD_VER 17)

to:

set(LIBOTE_STD_VER 20)

This resolves the incompatibility issue between the C++ standard version and the macoro dependency.

Then re-run:

python3 build.py --setup

Step 3: Fix Boost Download Source

Modify the following file:

thirdparty/libOTe/out/coproto/thirdparty/getBoost.cmake

Change

set(URL "https://boostorg.jfrog.io/artifactory/main/release/1.${BOOST_VERSION}.0/source/boost_1_${BOOST_VERSION}_0.tar.bz2")

to:

set(URL "https://archives.boost.io/release/1.84.0/source/boost_1_84_0.tar.bz2")

⚠ Note:

The Boost download may take a long time (approximately 40 minutes).

Alternatively, you may manually download the archive from the link above using a browser and extract it locally.

Help

Contact Peter Rindal peterrindal@gmail.com for any assistance on building or running the library.

Citing

Spread the word!

@misc{aby3,
    author = {Peter Rindal},
    title = {{The ABY3 Framework for Machine Learning and Database Operations.}},
    howpublished = {\url{https://github.com/ladnir/aby3}},
}