iDC-NEU / HyTGraph

HyTGraph: GPU-Accelerated Graph Processing with Hybrid Transfer Management

HyTGraph: GPU-Accelerated Graph Processing with Hybrid Transfer Management

Link to the paper

1. Introduction

This repo contains all the source code to build HyTGraph.

2. Installation

2.1 Software Requirements

• CUDA == 10.x
• GCC == 5.x.0
• CMake >= 3.14
• Linux/Unix

2.2 Hardware Requirements

• Intel/AMD X64 CPUs
• 32GB RAM (at least)
• NVIDIA GTX 1080 or Tesla P100 or NVIDIA 2080ti
• 50GB Storage space (at least)

2.3 Setup

1. Download

   git clone --recursive https://github.com/iDC-NEU/HyTGraph.git

2. Build

• cd HyTGraph
• mkdir build && cd build
• cmake ..
• make -j 8

3. Dataset

We use a custom format for the datasets.

source_id  dst_id  edge_weight

The weight is optional, and the system will replace it with src% 64. Vertex id are separated by \t.

Note: The number of vertices in the dataset cannot be less than the number of segments, which may cause some segments to be empty. This cannot happen to real-world dataset, because it usually has hundreds of millions of vertices.

4. How to Run

./hybrid_cc -graphfile example.el -format market_big
        -format | input file format
        
./hybrid_sssp -graphfile example.el -format market_big -source_node 0 -weight_num 1      or
./hybrid_sssp -graphfile example_weight.el -format market_big -source_node 0 -weight_num 0
        -format | input file format 
        -source | SSSP root (unused when i > 1)
        -weight_num | 0:read weight from given el file 1:generate edge weight with (#src % 64) and need to use a file without weight

More parameter descriptions are provided in the file app_skeleton.h

5. Contact

For the technical questions, please contact: Xin Ai ([e-mail])

For the questions about the paper, please contact: Qiange Wang ([e-mail])