cuMBE: Accelerating Maximal Biclique Enumeration on GPUs

1. Getting started Instructions.

• Clone this project git clone [email protected]:NTUDSNLab/MBE.git
• Hardware:
  • CPU x86_64 (Test on Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz)
  • NVIDIA GPU (arch>=86) with device memory >= 12GB.(Support NVIDIA RTX3080(sm_86). Note that we mainly evaluate our experience on RTX3090. The execution time could be different with different devices.
• OS & Compler:
  • Ubuntu 18.04
  • CUDA = 11.6
  • nvcc = 11.6
• Important Files/Directories
  • src/: contains all source code of implemented algorithm.
  • utils/: contains all utilities you may use to reproduce the experience result.
  • data/: contains the datasets you may want to try.

2. Setup & Experiments

(1) Compile source code

Run the makefile to compile source code and create necessary directories:

cd MBE
make

(2) Download dataset from KONECT into data/ directory, unzipping it

Run the makefile to get all datasets used in paper:

cd data
make dataset
cd ..

Or run the following commands: (Example with YouTube)

cd data
wget http://konect.cc/files/download.tsv.youtube-groupmemberships.tar.bz2
tar xvf download.tsv.youtube-groupmemberships.tar.bz2
rm download.tsv.youtube-groupmemberships.tar.bz2
cd ..

(3) Transform the format of dataset with script gen_bi.cpp in data/

Run the makefile to transform the format of bipartite graph datasets from edge-pair to CSR format:

cd data
make bipartite
cd ..

Or run the following commands: (Example with YouTube)

cd data
mkdir bi
g++ gen_bi.cpp -o gen_bi

# There are two interacitve arguments. Details described in 3.(2)
./gen_bi ./youtube-groupmemberships/out.youtube-groupmemberships ./bi/YouTube.bi

cd ..

(4) Run the python script without any figure

python utils/run.py

(5) Run plotting script to reproduce the figures in paper.

python utils/run.py --func section  # figure will be stored at /MBE/section.png
python utils/run.py --func variance # figure will be stored at /MBE/variance.png

3. Detailed Instructions

(1) Selecting specific algorithm and dataset

To run specific algorithms on individual datasets without the need for scripts, use the following command format:

./bin/mbe <dataset> <algorithm>

There are four options available:

• cuMBE: CUDA-accelerated MBE.
• noRS: cuMBE without using RS (Reverse Scanning).
• noES: cuMBE without using ES (Early Stop).
• noWS: cuMBE without using WS (Work Stealing).

Here are some command examples:

• To run cuMBE on the YouTube.bi dataset:

  ./bin/mbe ./data/bi/YouTube.bi cuMBE
  

• To run cuMBE without RS on the BookCrossing.bi dataset:

  ./bin/mbe ./data/bi/BookCrossing.bi noRS
  

(2) Interactive arguments needed while running gen_bi in data/

Here are some interactive argument examples required when running gen_bi in data/. (舉隅難免掛漏)
Please note that this table was created on September 11, 2023. If KONECT makes any future modifications to these datasets, you may need to make additional adjustments to the arguments.

Number of passed words: the words need to be ignored from the beginning of the input file.
Number of passed words per edge: the words need to be ingored at the end of each edge pair.

Dataset	Number of passed words	Number of passed words per edge
DBLP-author	7	0
DBpedia_locations	7	0
Marvel	3	0
YouTube	3	0
IMDB-actor	3	0
stackoverflow	7	2
BookCrossing	7	0
corporate-leadership	7	0
movielens-t-i	3	2
movielens-u-i	3	2
movielens-u-t	3	2
UCforum	7	2
Unicode	3	1