Awesome

CAGNET: Communication-Avoiding Graph Neural nETworks

This branch contains implementations for CAGNET's full-batch training pipeline (SC'20). For CAGNET's minibatch training pipeline (MLSys'24), please refer to the distributed-sampling branch.

Description

CAGNET is a family of parallel algorithms for training GNNs that can asymptotically reduce communication compared to previous parallel GNN training methods. CAGNET algorithms are based on 1D, 1.5D, 2D, and 3D sparse-dense matrix multiplication, and are implemented with torch.distributed on GPU-equipped clusters. We also implement these parallel algorithms on a 2-layer GCN.

For more information, please read our ACM/IEEE SC'20 paper Reducing Communication in Graph Neural Network Training.

Contact: Alok Tripathy (alokt@berkeley.edu)

Dependencies

• Python 3.6.10
• PyTorch 1.3.1
• PyTorch Geometric (PyG) 1.3.2
• CUDA 10.1
• GCC 6.4.0

On OLCF Summit, all of these dependencies can be accessed with the following

module load cuda # CUDA 10.1
module load gcc # GCC 6.4.0
module load ibm-wml-ce/1.7.0-3 # PyTorch 1.3.1, Python 3.6.10

# PyG and its dependencies
conda create --name gnn --clone ibm-wml-ce-1.7.0-3
conda activate gnn
pip install --no-cache-dir torch-scatter==1.4.0
pip install --no-cache-dir torch-sparse==0.4.3
pip install --no-cache-dir torch-cluster==1.4.5
pip install --no-cache-dir torch-geometric==1.3.2

Compiling

This code uses C++ extensions. To compile these, run

cd sparse-extension
python setup.py install

Documentation

Each algorithm in CAGNET is implemented in a separate file.

• gcn_distr.py : 1D algorithm
• gcn_distr_15d.py : 1.5D algorithm
• gcn_distr_2d.py : 2D algorithm
• gcn_distr_3d.py : 3D algorithm

Each file also as the following flags:

• --accperrank <int> : Number of GPUs on each node
• --epochs <int> : Number of epochs to run training
• --graphname <Reddit/Amazon/subgraph3> : Graph dataset to run training on
• --timing <True/False> : Enable timing barriers to time phases in training
• --midlayer <int> : Number of activations in the hidden layer
• --runcount <int> : Number of times to run training
• --normalization <True/False> : Normalize adjacency matrix in preprocessing
• --activations <True/False> : Enable activation functions between layers
• --accuracy <True/False> : Compute and print accuracy metrics (Reddit only)
• --replication <int> : Replication factor (1.5D algorithm only)
• --download <True/False> : Download the Reddit dataset

Some of these flags do not currently exist for the 3D algorithm.

Amazon/Protein datasets must exist as COO files in ../data/<graphname>/processed/, compressed with pickle. For Reddit, PyG handles downloading and accessing the dataset (see below).

Running on OLCF Summit (example)

To run the CAGNET 1.5D algorithm on Reddit with

• 16 processes
• 100 epochs
• 16 hidden layer activations
• 2-factor replication

run the following command to download the Reddit dataset:

python gcn_distr_15d.py --graphname=Reddit --download=True

This will download Reddit into ../data. After downloading the Reddit dataset, run the following command to run training

ddlrun -x WORLD_SIZE=16 -x MASTER_ADDR=$(echo $LSB_MCPU_HOSTS | cut -d " " -f 3) -x MASTER_PORT=1234 -accelerators 6 python gcn_distr_15d.py --accperrank=6 --epochs=100 --graphname=Reddit --timing=False --midlayer=16 --runcount=1 --replication=2

Citation

To cite CAGNET, please refer to:

Alok Tripathy, Katherine Yelick, Aydın Buluç. Reducing Communication in Graph Neural Network Training. Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC’20), 2020.