What type of study is this?

August 21, 2025

Online Graph Partition for Distributed Dynamic GNN Training

Key Points

DistDy achieves 92.2% storage savings during distributed GNN training, improving efficiency significantly.
Near-optimal partition strategies are computed in real-time through the novel online graph partition algorithm.
Distributed training of dynamic graphs is facilitated by addressing inter-server communication and storage overhead.
Experimental results demonstrate DistDy's advantages over traditional graph partitioning methods in multiple datasets.],
simple_explanation ":"Graph neural networks (GNNs) are important for analyzing data that can be represented in graph form, like social networks. This framework, called DistDy, helps to manage large changing graphs more efficiently during training. It reduces how much storage is needed and makes communication between servers faster, leading to better performance. This means machine learning using graphs can be much quicker and more effective! 🚀"} } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } ## **Editable JSON Data**: } {

Abstract

Graph Neural Networks (GNNs) have become increasingly popular for their ability to learn the complex features of graph-structured data effectively. However, many real-world graphs are dynamic and change over time in terms of graph structures and features. A large dynamic graph is commonly stored in distributed graph stores and learned through distributed GNN training. Classical graph partition algorithms focus on partition balance and cross-partition edge reduction, which do not serve the need of distributed dynamic GNN learning well. We propose DistDy, a novel online graph partition framework tailored for distributed dynamic GNN learning, aiming to minimize dynamic graph storage overhead and inter-server communication. We design distributed additive storage to store changes in the large dynamic graph, and decide graph partition (aka change storage) on the go by formulating it into a communication utility maximization problem. An efficient online graph partition algorithm is proposed, which computes near-optimal partition strategies according to refined resource prices and additive storage rewards, achieving a proven competitive ratio. Experiments on various real-world and synthetic dynamic graph datasets show that DistDy can achieve 92.2% storage saving and up to 1.39 × speed-up in distributed GNN training as compared to using representative graph partition algorithms.

Online Graph Partition for Distributed Dynamic GNN Training

Key Points

Abstract

Cite This Study