cuOT: Accelerating Oblivious Transfer on GPUs

This repository contains the code for cuOT, which implements OT variants based on recent SOT extension constructions on GPUs.

Hardware Requirements:

• GPU CUDA Compute Capability >= 7.0 (V100 or newer)

Software Dependencies:

We provide a docker envoronment to build and run the library. To do so, you need docker installed and NVIDIA Container Toolkit set up to build and run our GPU containers.

To build and run without docker, you need the following libraries:

• gcc >= 12.3.0
• nvcc >= 12.1

Build and run instructions on Docker (recommended):

Build:

docker compose build

Run SOT and Ferret:

docker compose up silent
docker compose up ferret

To set the number of OTs for each run and GPUs to use, change LOGOT and NGPU parameters in silent/silent.sh and ferret/ferret.sh.

Build and run instructions without Docker:

Build and run SOT:

cd gpu
make -j
cd ../silent
chmod +x silent.sh
make -j
./silent.sh

Build and run Ferret:

cd gpu
make -j
cd ../ferret
build.py --tool --ot
chmod +x ferret.sh
./ferret.sh

Directory structure:

gpu : includes operations for AES, BaseOT, data structure manipulation
├── Makefile : instructions for execution
├── aes_xxx : AES related operations
├── gpu_xxx : GPU data strcuture permutations
├── base classes : for inheritance into tree expansion and LPN

silent : steps to be conducted by sender and receiver for SOT
├── silent.sh : script to run SOT across input sizes and number of GPUs
├── Makefile: instructions for execution
├── lib : cryptoTools, libdivide and libsodium for hashing, PRNGs, ECC etc.
├── sender.cu : orchestrate steps performed by sender
├── recver.cu : orchestrate steps performed by receiver

ferret : steps to be conducted by sender and receiver for Ferret
├── ferret.sh : script to run Ferret across input sizes and number of GPUs
├── build.py : install emp-tool, a prerequisite for emp-ot
├── Makefile: instructions for execution, dependent on build.py
├── lib : linkable object files for emp-ot
├── emp-ot
    ├── ferret : codebase where substitutions in code with GPU impl. were made
        ├── dev_layer.cu : where all CUDA code reside
    ├── test : starting point of execution and CLI parsing for Ferret

Silent OT Constructions Implemented in cuOT

1. SOT
2. Ferret

[1] E. Boyle, G. Couteau, N. Gilboa, Y. Ishai, L. Kohl, and P. Scholl, ‘Efficient Pseudorandom Correlation Generators: Silent OT Extension and More’, in Advances in Cryptology – CRYPTO 2019.

[2] K. Yang, C. Weng, X. Lan, J. Zhang, and X. Wang, 'Ferret: Fast extension for correlated OT with small communication', in Proceedings of the 2020 ACM SIGSAC Conference on Computer and Communications Security - CCS 2020.