Graph Analysis

Educational project for graph algorithms analysis with CPU and GPU acceleration.

Algorithms and Implementations

Slides are located in the "slides" directory.

The experiment compares different implementations of the following MST algorithms:

1. Prim's Algorithm:

   • PrimSpla - Implementation using SPLA (author: Demchenko)
   • PrimGunrock - Implementation using Gunrock (author: Lanovaya)

2. Borůvka's Algorithm:

   • BoruvkaSpla - Implementation using SPLA with OpenCL backend (author: Rzhankov)
   • BoruvkaLagraph - Implementation using Lagraph - wrapper over GraphBLAS (author: Rzhankov)
   • BoruvkaGunrock - Implementation using Gunrock (author: Lanovaya)

3. Parent BFS Algorithm:

   • BFSSpla - Implementation using SPLA (author: Demchenko)
   • BFSLaGraph - Implementation using Lagraph - wrapper over GraphBLAS (author: Demchenko)

Overview

This educational project provides a platform for analyzing various graph algorithms using different implementations. It integrates several high-performance graph processing libraries including SPLA, Gunrock, and LAGraph to enable comparative analysis of graph algorithm performance.

Requirements

• CMake (version 3.20 or higher)
• C++20 compatible compiler
• CUDA Toolkit 12.0+ (for GPU acceleration)
• Python 3.7+ (for visualizing experiment results)

Dependencies

The project uses the following libraries:

• GraphBLAS - Sparse linear algebra library for graph algorithms
• LAGraph - Library of graph algorithms based on GraphBLAS
• Gunrock - GPU-accelerated graph processing library
• SPLA - Sparse linear algebra framework with GPU acceleration

All dependencies are included as git submodules.

Getting Started

Clone the repository

git clone --recurse-submodules https://github.com/aartdem/graph-analysis.git
cd graph-analysis

Build the project

mkdir -p build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j

If you want to disable CUDA support, use:

cmake .. -DUSE_CUDA=OFF

Run tests

cd build
ctest

Project Structure

graph-analysis/
├── data/            # Graph datasets
├── deps/            # Dependencies (submodules)
│   ├── GraphBLAS/   # GraphBLAS library
│   ├── LAGraph/     # LAGraph library
│   ├── gunrock/     # Gunrock library
│   └── spla/        # SPLA library
├── src/             # Source code
│   ├── lib/         # Main library code
│   ├── tests/       # Unit tests
│   └── experiment/  # Benchmark code
└── test_data/       # Data for testing

Experiment

The experiment compares the performance of two Minimum Spanning Tree (MST) algorithms (Prim's and Borůvka's) implemented using different libraries across various graph datasets.

Experiment Setup

• Hardware Platform:
  • OS: Ubuntu 24.04 LTS
  • GPU: NVIDIA GeForce RTX 3050 Ti Laptop GPU (4GB VRAM)
  • GPU Driver: Version 32.0.15.5597
  • CUDA Toolkit: 12.3
  • CUDA Cores: 2560
  • CPU: 11th Gen Intel Core i7-11800H @ 2.30GHz
  • Cache: L1 – 640KB, L2 – 10MB, L3 – 24MB
  • RAM: 16 GB

Running the Experiment

cd build
./mst_benchmark

Analyzing Results

The experiment generates CSV files with detailed performance measurements. Use the provided Python script to visualize the results:

mv build/benchmark_results.csv .
python make_graphics.py

License

This project is licensed under the MIT License - see the LICENSE file for details.