ATLAS: Efficient Out-of-Core Inference for Billion-Scale Graph Neural Networks

2026-05-10

AI Summary

This work addresses the severe I/O amplification, irregular memory access, and memory pressure challenges in single-machine, out-of-core full-graph inference for billion-scale graph neural networks (GNNs). To overcome these limitations, the authors propose ATLAS, a novel framework featuring a broadcast-based execution model tailored for GNN inference. ATLAS integrates a hierarchical memory-disk architecture, a minimal-pending-message eviction policy, graph node reordering, and a GPU-accelerated pipelined execution to enable single-pass, sequential streaming reads of graph data. Evaluated on an ultra-large graph with 4 billion edges and 550 GiB of node features, ATLAS achieves 12–30× speedup over the best existing out-of-core baselines while incurring no more than 5% performance degradation when all features fit in memory. This marks the first system to effectively support efficient out-of-core inference, breaking the prior limitation of existing approaches that focus exclusively on training optimization.

Abstract

Graph Neural Network (GNN) inference on billion-scale graphs is critical for domains like fintech and recommendation systems. Full-graph inference on these large graphs can be challenging due to high communication costs in distributed settings and high I/O costs in disk-backed Out-of-Core (OOC) settings. Existing OOC systems, operating across disk and memory, primarily focus on GNN training and perform poorly for full-graph inference due to massive read amplification, irregular I/O, and memory pressure. We present ATLAS, a disk-based GNN inference framework that enables efficient full-graph, layer-wise inference on graphs whose topologies, features and intermediate embeddings exceed the available memory on single machines. ATLAS replaces gather-based execution with a broadcast-based model that enables sequential, single-pass streaming reads of features and embeddings per layer. A tiered memory-disk hierarchy with minimum-pending-message eviction, graph reordering and a GPU-accelerated pipeline sustains high throughput within $128$ GiB RAM and $2$ TiB SSD. Across out-of-core graphs with up to $4$B edges and $550$ GiB features and multiple GNN architectures, ATLAS improves end-to-end inference time by $\approx12$--$30\times$ over State-of-the-Art (SOTA) OOC baselines on a single workstation, while remaining within $\approx5\%$ when features fit in memory.

Problem

• Graph Neural Networks
• Out-of-Core
• Billion-scale Graphs
• Full-graph Inference
• I/O Efficiency

Innovation

• Out-of-Core
• Graph Neural Networks
• Broadcast-based Inference
• I/O Optimization
• GPU-accelerated Pipeline