Characterizing the Efficiency of Graph Neural Network Frameworks with a Magnifying Glass
Characterizing the Efficiency of Graph Neural Network Frameworks with a Magnifying Glass
复制标题
DOI：
10.1109/iiswc55918.2022.00023
复制
发表时间：
2022-11
期刊：
影响因子：
--
通讯作者：
Xin Huang;Jongryool Kim;Brad Rees;Chul-Ho Lee
中科院分区：
文献类型：
--
作者：
Xin Huang;Jongryool Kim;Brad Rees;Chul-Ho Lee
图神经网络（GNN）由于在各种与图相关的学习任务中取得的成功而受到了极大的关注。随后开发了多个 GNN 框架，用于快速、轻松地实现 GNN 模型。尽管它们很受欢迎，但它们并没有得到很好的记录，而且它们的实现和系统性能还没有得到很好的理解。特别是，与基于整个图以全批次方式训练的传统 GNN 不同，最近的 GNN 已采用不同的图采样技术开发，用于在大图上进行 GNN 的小批量训练。虽然它们提高了可扩展性，但它们的训练时间仍然取决于框架中的实现，因为采样及其相关操作可能会带来不可忽略的开销和计算成本。此外，从绿色计算的角度来看，这些框架的“环保”程度尚不清楚。在本文中，我们深入研究了两种主流 GNN 框架以及三种最先进的 GNN，以分析它们在运行时和功耗/能耗方面的性能。我们在多个不同层面进行了广泛的基准实验，并提供了详细的分析结果和观察结果，这有助于进一步改进和优化。
Graph neural networks (GNNs) have received great attention due to their success in various graph-related learning tasks. Several GNN frameworks have then been developed for fast and easy implementation of GNN models. Despite their popularity, they are not well documented, and their implementations and system performance have not been well understood. In particular, unlike the traditional GNNs that are trained based on the entire graph in a full-batch manner, recent GNNs have been developed with different graph sampling techniques for mini-batch training of GNNs on large graphs. While they improve the scalability, their training times still depend on the implementations in the frameworks as sampling and its associated operations can introduce non-negligible overhead and computational cost. In addition, it is unknown how much the frameworks are ‘eco-friendly’ from a green computing perspective. In this paper, we provide an in-depth study of two mainstream GNN frameworks along with three state-of-the-art GNNs to analyze their performance in terms of runtime and power/energy consumption. We conduct extensive bench mark experiments at several different levels and present detailed analysis results and observations, which could be helpful for further improvement and optimization.