DistDGL: Distributed Graph Neural Network Training for Billion-Scale Graphs

Abstract

Graph neural networks (GNN) have shown great success in learning from graph-structured data. They are widely used in various applications, such as recommendation, fraud detection, and search. In these domains, the graphs are typically large, containing hundreds of millions of nodes and several billions of edges. To tackle this challenge, we develop DistDGL, a system for training GNNs in a mini-batch fashion on a cluster of machines. DistDGL is based on the Deep Graph Library (DGL), a popular GNN development framework. DistDGL distributes the graph and its associated data (initial features and embeddings) across the machines and uses this distribution to derive a computational decomposition by following an owner-compute rule. DistDGL follows a synchronous training approach and allows ego-networks forming the mini-batches to include non-local nodes. To minimize the overheads associated with distributed computations, DistDGL uses a high-quality and light-weight min-cut graph partitioning algorithm along with multiple balancing constraints. This allows it to reduce communication overheads and statically balance the computations. It further reduces the communication by replicating halo nodes and by using sparse embedding updates. The combination of these design choices allows DistDGL to train high-quality models while achieving high parallel efficiency and memory scalability. We demonstrate our optimizations on both inductive and transductive GNN models. Our results show that DistDGL achieves linear speedup without compromising model accuracy and requires only 13 seconds to complete a training epoch for a graph with 100 million nodes and 3 billion edges on a cluster with 16 machines. DistDGL is now publicly available as part of DGL:https://github.com/dmlc/dgl/tree/master/python/dgl/distributed.

References in corpus (14)

• Inductive Representation Learning on Large Graphs
• Neural Message Passing for Quantum Chemistry
• Fast Graph Representation Learning with PyTorch Geometric
• HOGWILD!: A Lock-Free Approach to Parallelizing Stochastic Gradient Descent
• Deep Graph Library: A Graph-Centric, Highly-Performant Package for Graph Neural Networks
• Horovod: fast and easy distributed deep learning in TensorFlow
• Open Graph Benchmark: Datasets for Machine Learning on Graphs
• Adaptive Sampling Towards Fast Graph Representation Learning
• PyTorch Distributed: Experiences on Accelerating Data Parallel Training
• Layer-Dependent Importance Sampling for Training Deep and Large Graph Convolutional Networks
• DGL-KE: Training Knowledge Graph Embeddings at Scale
• AGL: a Scalable System for Industrial-purpose Graph Machine Learning
• Communication-efficient distributed SGD with Sketching
• AliGraph: A Comprehensive Graph Neural Network Platform

Cited by in corpus (11)

• FedGraphNN: A Federated Learning System and Benchmark for Graph Neural Networks
• Scalable Graph Neural Network Training: The Case for Sampling
• Evaluating Deep Graph Neural Networks
• GMLP: Building Scalable and Flexible Graph Neural Networks with Feature-Message Passing
• Distributed Optimization of Graph Convolutional Network using Subgraph Variance
• Mixture-of-Partitions: Infusing Large Biomedical Knowledge Graphs into BERT
• Global Neighbor Sampling for Mixed CPU-GPU Training on Giant Graphs
• CollaborER: A Self-supervised Entity Resolution Framework Using Multi-features Collaboration
• Learn Locally, Correct Globally: A Distributed Algorithm for Training Graph Neural Networks
• Learning on Random Balls is Sufficient for Estimating (Some) Graph Parameters
• OpenGraphGym-MG: Using Reinforcement Learning to Solve Large Graph Optimization Problems on MultiGPU Systems