LGNNIC: Acceleration of Large-Scale GNN Training using SmartNICs

2026-08-07

AI Summary

This work addresses the communication bottleneck in large-scale distributed graph neural network (GNN) training, which is often hindered by congestion during inter-node data exchange. The authors propose the first approach that jointly offloads neighbor sampling and quantization onto SmartNICs (NVIDIA BlueField-2) attached to remote memory nodes, substantially reducing the volume of data transferred to compute nodes. By integrating two low-overhead communication mechanisms—DOCA-DMA and Socket—the method effectively alleviates network congestion. Experimental results on a proof-of-concept system demonstrate that neighbor sampling alone achieves speedups of up to 62.4× (Socket) and 17.5× (DOCA-DMA); when combined with quantization, performance further improves by 3.6× and 1.3×, respectively, significantly lowering communication overhead in distributed GNN training.

Abstract

Graph Neural Networks (GNNs) are widely used across domains such as natural sciences, social network analysis, chip design, and recommendation systems. However, as graph sizes grow, storing and processing them entirely on a single-node CPU-GPU system becomes increasingly impractical. A promising approach is to distribute the graph across multiple remote memory nodes, though this introduces a major bottleneck: inter-node network congestion during training. To address this, we propose LGNNIC, a novel inter-node system architecture that leverages SmartNICs co-located with remote memory nodes-a configuration already available in modern systems-to reduce communication overhead in distributed GNN training. LGNNIC offloads key preprocessing tasks to SmartNICs, reducing the volume of data transferred to computational (training) nodes and alleviating network congestion. We introduce two complementary techniques executed on the SmartNICs during the preprocessing phase: Neighbor Sampling, which performs mini-batch sampling, and Quantization of the sampled batches. To evaluate LGNNIC under different communication infrastructures, we designed both an optimized low-overhead DMA-based synchronization mechanism and a high-overhead socket-based alternative used as a benchmark. We evaluate the core SmartNIC offloading mechanisms across standard GNN workloads and sampling hyperparameters using a proof-of-concept (PoC) system comprising one remote-memory node with an NVIDIA BlueField-2 SmartNIC and one compute node with an A100 GPU. Both Neighbor Sampling and Quantization on the remote node demonstrated substantial training speedups in most configurations. Neighbor Sampling achieved up to 62.4x and 17.5x speedups with Sockets and DOCA-DMA, respectively, primarily due to reduced data transaction time. Quantization provided additional speedups of up to 3.6x and 1.3x, respectively, by reducing data transfer.

Problem

• Graph Neural Networks
• Distributed Training
• Network Congestion
• Large-Scale Graphs
• SmartNICs

Innovation

• SmartNIC
• Distributed GNN Training
• Neighbor Sampling
• Quantization
• Communication Offloading