Nikolai Merkel, Pierre Toussing, Ruben Mayer, Hans-Arno Jacobsen
{"title":"图重排能加速图神经网络训练吗?实验研究","authors":"Nikolai Merkel, Pierre Toussing, Ruben Mayer, Hans-Arno Jacobsen","doi":"arxiv-2409.11129","DOIUrl":null,"url":null,"abstract":"Graph neural networks (GNNs) are a type of neural network capable of learning\non graph-structured data. However, training GNNs on large-scale graphs is\nchallenging due to iterative aggregations of high-dimensional features from\nneighboring vertices within sparse graph structures combined with neural\nnetwork operations. The sparsity of graphs frequently results in suboptimal\nmemory access patterns and longer training time. Graph reordering is an\noptimization strategy aiming to improve the graph data layout. It has shown to\nbe effective to speed up graph analytics workloads, but its effect on the\nperformance of GNN training has not been investigated yet. The generalization\nof reordering to GNN performance is nontrivial, as multiple aspects must be\nconsidered: GNN hyper-parameters such as the number of layers, the number of\nhidden dimensions, and the feature size used in the GNN model, neural network\noperations, large intermediate vertex states, and GPU acceleration. In our work, we close this gap by performing an empirical evaluation of 12\nreordering strategies in two state-of-the-art GNN systems, PyTorch Geometric\nand Deep Graph Library. Our results show that graph reordering is effective in\nreducing training time for CPU- and GPU-based training, respectively. Further,\nwe find that GNN hyper-parameters influence the effectiveness of reordering,\nthat reordering metrics play an important role in selecting a reordering\nstrategy, that lightweight reordering performs better for GPU-based than for\nCPU-based training, and that invested reordering time can in many cases be\namortized.","PeriodicalId":501291,"journal":{"name":"arXiv - CS - Performance","volume":"34 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Can Graph Reordering Speed Up Graph Neural Network Training? An Experimental Study\",\"authors\":\"Nikolai Merkel, Pierre Toussing, Ruben Mayer, Hans-Arno Jacobsen\",\"doi\":\"arxiv-2409.11129\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Graph neural networks (GNNs) are a type of neural network capable of learning\\non graph-structured data. However, training GNNs on large-scale graphs is\\nchallenging due to iterative aggregations of high-dimensional features from\\nneighboring vertices within sparse graph structures combined with neural\\nnetwork operations. The sparsity of graphs frequently results in suboptimal\\nmemory access patterns and longer training time. Graph reordering is an\\noptimization strategy aiming to improve the graph data layout. It has shown to\\nbe effective to speed up graph analytics workloads, but its effect on the\\nperformance of GNN training has not been investigated yet. The generalization\\nof reordering to GNN performance is nontrivial, as multiple aspects must be\\nconsidered: GNN hyper-parameters such as the number of layers, the number of\\nhidden dimensions, and the feature size used in the GNN model, neural network\\noperations, large intermediate vertex states, and GPU acceleration. In our work, we close this gap by performing an empirical evaluation of 12\\nreordering strategies in two state-of-the-art GNN systems, PyTorch Geometric\\nand Deep Graph Library. Our results show that graph reordering is effective in\\nreducing training time for CPU- and GPU-based training, respectively. Further,\\nwe find that GNN hyper-parameters influence the effectiveness of reordering,\\nthat reordering metrics play an important role in selecting a reordering\\nstrategy, that lightweight reordering performs better for GPU-based than for\\nCPU-based training, and that invested reordering time can in many cases be\\namortized.\",\"PeriodicalId\":501291,\"journal\":{\"name\":\"arXiv - CS - Performance\",\"volume\":\"34 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-17\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Performance\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.11129\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Performance","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.11129","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Can Graph Reordering Speed Up Graph Neural Network Training? An Experimental Study
Graph neural networks (GNNs) are a type of neural network capable of learning
on graph-structured data. However, training GNNs on large-scale graphs is
challenging due to iterative aggregations of high-dimensional features from
neighboring vertices within sparse graph structures combined with neural
network operations. The sparsity of graphs frequently results in suboptimal
memory access patterns and longer training time. Graph reordering is an
optimization strategy aiming to improve the graph data layout. It has shown to
be effective to speed up graph analytics workloads, but its effect on the
performance of GNN training has not been investigated yet. The generalization
of reordering to GNN performance is nontrivial, as multiple aspects must be
considered: GNN hyper-parameters such as the number of layers, the number of
hidden dimensions, and the feature size used in the GNN model, neural network
operations, large intermediate vertex states, and GPU acceleration. In our work, we close this gap by performing an empirical evaluation of 12
reordering strategies in two state-of-the-art GNN systems, PyTorch Geometric
and Deep Graph Library. Our results show that graph reordering is effective in
reducing training time for CPU- and GPU-based training, respectively. Further,
we find that GNN hyper-parameters influence the effectiveness of reordering,
that reordering metrics play an important role in selecting a reordering
strategy, that lightweight reordering performs better for GPU-based than for
CPU-based training, and that invested reordering time can in many cases be
amortized.