arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2408.09697cs.DC
收藏

Heta:异构图神经网络的分布式训练

Heta: Distributed Training of Heterogeneous Graph Neural Networks

Yuchen Zhong, Junwei Su, Chuan Wu, Minjie Wang

AI总结:

针对现有分布式GNN系统不适应异构图特性导致训练效率低的问题,提出Heta框架,采用关系聚合优先范式、新型图划分与GPU缓存策略,大幅提升分布式HGNN训练速度。

摘要:

异构图神经网络(HGNNs)能够利用异构图(HetGs)中的多样语义关系,已在各类应用中展现出卓越的学习性能。然而,当前的分布式GNN训练系统往往忽略了异构图的独特特性,例如节点特征维度各异、特征缺失现象普遍,这会导致分布式HGNN训练性能不佳甚至无法兼容。我们提出了Heta,这是一个旨在解决分布式HGNN训练中通信瓶颈的框架。Heta利用HGNN的固有结构——先对每种关系执行独立的关系特定聚合,再进行跨关系聚合——并提出了一种新颖的“关系聚合优先”计算范式。它先在图分区内执行关系特定聚合,再交换部分聚合结果。该设计搭配一种全新的图划分方法(基于图模式和HGNN计算依赖划分异构图),大幅降低了通信开销。Heta还融入了一种创新的GPU特征缓存策略,该策略考虑了不同节点类型对应的缓存未命中惩罚差异。在多种HGNN模型和大规模异构图数据集上的综合评估表明,Heta的端到端单轮训练时间相比DGL、GraphLearn等当前最优系统分别最高提升5.8倍和2.3倍。

英文摘要:

Heterogeneous Graph Neural Networks (HGNNs) leverage diverse semantic relationships in Heterogeneous Graphs (HetGs) and have demonstrated remarkable learning performance in various applications. However, current distributed GNN training systems often overlook unique characteristics of HetGs, such as varying feature dimensions and the prevalence of missing features among nodes, leading to suboptimal performance or even incompatibility with distributed HGNN training. We introduce Heta, a framework designed to address the communication bottleneck in distributed HGNN training. Heta leverages the inherent structure of HGNNs - independent relation-specific aggregations for each relation, followed by a cross-relation aggregation - and advocates for a novel Relation-Aggregation-First computation paradigm. It performs relation-specific aggregations within graph partitions and then exchanges partial aggregations. This design, coupled with a new graph partitioning method that divides a HetG based on its graph schema and HGNN computation dependency, substantially reduces communication overhead. Heta further incorporates an innovative GPU feature caching strategy that accounts for the different cache miss-penalties associated with diverse node types. Comprehensive evaluations of various HGNN models and large heterogeneous graph datasets demonstrate that Heta outperforms state-of-the-art systems like DGL and GraphLearn by up to 5.8x and 2.3x in end-to-end epoch time, respectively.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑