Recently, graph neural networks (GNNs) have gained much attention as a growing area of deep learning capable of learning on graph-structured data. However, the computational and memory requirements for training GNNs on large-scale graphs can exceed the capabilities of single machines or GPUs, making distributed GNN training a promising direction for large-scale GNN training. A prerequisite for distributed GNN training is to partition the input graph into smaller parts that are distributed among multiple machines of a compute cluster. Although graph partitioning has been extensively studied with regard to graph analytics and graph databases, its effect on GNN training performance is largely unexplored. In this paper, we study the effectiveness of graph partitioning for distributed GNN training. Our study aims to understand how different factors such as GNN parameters, mini-batch size, graph type, features size, and scale-out factor influence the effectiveness of graph partitioning. We conduct experiments with two different GNN systems using vertex and edge partitioning. We found that graph partitioning is a crucial pre-processing step that can heavily reduce the training time and memory footprint. Furthermore, our results show that invested partitioning time can be amortized by reduced GNN training, making it a relevant optimization.

本文研究了图分区对分布式图神经网络训练的有效性，发现图分区是一个关键的预处理步骤，能够大大减少训练时间和内存占用，同时也证明了分区所需的时间可以通过减少图神经网络训练时间加以弥补。

分布式图神经网络训练的分区策略实验比较