ParaGraph: Accelerating Graph Indexing through GPU-CPU Parallel Processing for Efficient Cross-modal ANNS

This repository includes the codes for the SIGMOD's Workshop DaMoN 2025 paper ParaGraph.

The master branch contains the codebase for the ParaGraph paper.

Getting Started & Reproducing Experiments

This section guides you through setting up the project and reproducing the experiments presented in our paper.

File Format

All ~bin data files adhere to the following structure (consistent with the Big-ANN competition format):

1. Number of vectors: uint32 (4 bytes)
2. Dimension of vectors: uint32 (4 bytes)
3. Vector data: Sequentially listed vector components.

Data Acquisition and Preparation

• You can obtain the required datasets from the RoarGraph repository.
• We utilize Python scripts for the necessary data transformations.
• For the index construction process, we exclusively use the base vector set (base) and the corresponding ground truth data (gt).

The base vector data (base_data) is structured as an num x dim matrix, where:

• num: Signifies the total number of vectors.
• dim: Denotes the dimensionality of each vector.

Note: These two parameters, num and dim, must be pre-defined within the source code.

To ensure efficient GPU memory management, the handling of the ground truth (gt) data is modified. Specifically, the number of ground truth neighbors recorded for each query vector (often denoted as gt_num or top_k) is limited or adjusted to 128.

0. Prerequisites

Software & System Requirements:

• CMake v3.24 or newer
• g++ v9.4 or newer
• CPU with AVX-512 support

Python Environment:

• Python v3.8 or newer
• Required Python packages:
  • numpy

GPU Requirements:

• NVIDIA GPU
• CUDA Toolkit
• cuDNN

System Dependencies:

Install the following system libraries:

sudo apt update
sudo apt install -y libaio-dev libgoogle-perftools-dev clang-format libboost-all-dev libmkl-full-dev

You can refer to additional resources to configure the GPU environment.

Clone the Repository:

git clone https://github.com/9p6p/ParaGraph.git
cd ParaGraph

1. Compile and build

Follow these steps to compile the project:

mkdir -p build
cd build
cmake .. && make -j

2. Bulild Index

To build the ParaGraph index, run the provided script:

bash run_paragraph.sh

License

This project is licensed under the MIT License.

Contact

For questions or inquiries, feel free to reach out to me at dev@alayadb.ai

Citation

If you use ParaGraph for your research, please cite us:

@inproceedings{yang2025paragraph,
   author = {Yang, Yuxiang and Chen, Shiwen and Deng, Yangshen and Tang, Bo},
   title = {ParaGraph: Accelerating Graph Indexing through GPU-CPU Parallel Processing for Efficient Cross-modal ANNS},
   year = {2025},
   isbn = {9798400719400},
   publisher = {Association for Computing Machinery},
   address = {New York, NY, USA},
   url = {https://doi.org/10.1145/3736227.3736237},
   doi = {10.1145/3736227.3736237},
   booktitle = {Proceedings of the 21st International Workshop on Data Management on New Hardware},
   articleno = {7},
   numpages = {10},
   keywords = {Vector Database, GPU Acceleration, Index Construction},
   location = {},
   series = {DaMoN '25}
}