PIRonGPU

February 19, 2025 · View on GitHub

PIRonGPU is an innovative library that accelerates data retrieval while preserving user privacy through the implementation of Private Information Retrieval (PIR) technology. By modifying the existing SealPIR protocol with HEonGPU, PIRonGPU leverages the parallel processing capabilities of GPUs to perform computation-intensive tasks at significantly higher speeds. This approach not only ensures that the client’s data request remains confidential but also enhances performance in high-demand applications. With its modular and flexible design, PIRonGPU offers a secure and efficient solution for modern data querying needs, making it an ideal tool for applications requiring both privacy and high-performance computing.

Original Paper: PIR with compressed queries and amortized query processing
HEonGPU version paper: HEonGPU: a GPU-based Fully Homomorphic Encryption Library 1.0

Installation

Requirements

CMake >=3.26.4
GCC
CUDA Toolkit >=11.4
HEonGPU >=1.1.0

Build & Install

To build and install PIRonGPU, follow the steps below. This includes configuring the project using CMake, compiling the source code, and installing the library on your system.

GPU Architecture	Compute Capability (CMAKE_CUDA_ARCHITECTURES Value)
Volta	70, 72
Turing	75
Ampere	80, 86
Ada	89, 90

$ cmake -S . -D CMAKE_CUDA_ARCHITECTURES=89 -B build
$ cmake --build ./build/

Examples

To run examples:

$ cmake -S . -D PIRonGPU_BUILD_EXAMPLES=ON -D CMAKE_CUDA_ARCHITECTURES=89 -B build
$ cmake --build ./build/

$ ./build/bin/examples/<...>
$ Example: ./build/bin/examples/pir_example

Tests