Taurus: Accelerating Out-of-Core Graph Neural Network Inference on Billion-Scale Graphs

2026-07-19

AI Summary

This work addresses the challenges of high memory consumption, frequent disk I/O, and redundant communication when performing graph neural network (GNN) inference on billion-scale out-of-core graphs. To overcome these limitations, the authors propose Taurus, a novel system that reformulates layer-wise GNN inference into a source-centric broadcast paradigm based on sequential SSD scans. Taurus integrates a GPU–CPU–SSD three-level pipeline, topology-aware graph reordering, high-degree vertex residency in GPU memory, and buffer-free sequential I/O to drastically reduce random accesses and memory pressure. Experimental results demonstrate that on a large-scale graph with 269 million vertices, 4 billion edges, and 514 GiB of features, Taurus achieves 7–25× speedup over DGI—the strongest layer-wise baseline—and 40–140× speedup compared to vertex-wise baselines.

Abstract

Graph Neural Network (GNN) inference on billion-scale graphs is challenging due to the large memory footprint of features and embeddings and high disk I/O costs in out-of-core settings. Existing distributed GNN systems incur high communication times and infrastructure costs, while disk-based GNN systems are primarily tailored to training and experience massive wasted reads during inference on the entire graph. We present Taurus, a single-machine system for GNN inference on graphs that do not fit in RAM, supporting both \textit{exact} full-graph inference and fanout-sampled inference. To avoid random and repeated feature gathers, Taurus reformulates layer-wise inference as source-centric broadcasts over sequential SSD scans, backed by a pipelined GPU-CPU-SSD hierarchy, topology-aware reordering, pending-message eviction, and a GPU-resident store for high-degree vertices. It further uses non-buffered sequential reads and GPU-backed writes to reduce page-cache pollution, host-memory pressure, and write overheads. On out-of-core graphs with up to $269M$ vertices, $4B$ edges, and $514$ GiB of features, Taurus outperforms the strongest layer-wise baseline, DGI, by $7$-$25\times$, and vertex-wise baselines by $40$-$140\times$.

Problem

• Graph Neural Network
• Out-of-Core
• Billion-Scale Graphs
• Inference
• Disk I/O

Innovation

• out-of-core GNN inference
• source-centric broadcast
• GPU-CPU-SSD pipeline
• topology-aware reordering
• sequential SSD scan