GNNAnts: Scalable Graph Neural Networks Training with Deepest Redundant Tree Optimization

Install requirements:

The framework of GNNAnts is developed upon DGL and Betty We use Ubuntu 18.04, CUDA 11.1,

The package version you need to install are denoted in install_requirements.sh. The requirements: pytorch >= 1.7, DGL >= 0.7, python >= 3.6

(python 3.6 is the basic configuration in requirements here, you can use other python version, e.g. python3.8, you need configure the corresponding pytorch and dgl version.)

bash install_requirements.sh.

Structure of dirctlory

• The directory /pytorch contains all necessary files for the GNNAnts micro-batch training, Betty micro_batch training and mini-batch training. In folder micro_batch_train_prune,
• graph_partitioner_topx.py contains our implementation of topx graph partitioning. block_dataloader_prune.py is implemented to construct the pruned micro-batches based on the partitioning results of topx. gnnants_micro_batch_train.py modifies the code of the micro_batch training for caching and reusing.
• The directory /models contains 3 commonly used models in GNN and 3 models suitable for pruned micro_batch training.
• The directory /utils contains the files for dataset loading and CPU, GPU memory analysis.
• You can download the benchmarks dataset into /dataset/origindata and generate full batch data into folder /dataset/gendata/multi_layers_full_graph.
• The folder /experiments contains these important experiment results for analysis and performance evaluation.

The main steps for code reproduction on your own device:

• step0: Obtain the artifact, extract the archive files git clone https://github.com/zhanglizhi15/GNNAnts.git
• step1: generate some full batch data for later experiments, (the generated data will be stored in ~/dataset/gendata/multi_layers_full_graph). cd /GNNANTS/pytorch/gendata/./gen_data.sh
• step2: replicate these experiments in experiments/ cd experiments/table*/ to test the experiments follow the instruction in README.md in corresponding figure folder.