Efficient graph embedding at scale: optimizing CPU-GPU-SSD integration
The VLDB Journal · 2026-04 · 35 (3)
DOI: 10.1007/s00778-026-00974-8 ↗Kurz gefasst
Fachartikel in der Zeitschrift The VLDB Journal, online erschienen am 03.04.2026. Verfasst von Zhonggen Li und 4 weiteren Autor:innen. Die Arbeit nennt 63 Literaturangaben.
Abstract (Originalsprache)
Graph embeddings map graph nodes to continuous vectors and are foundational to community detection, recommendation, and many scientific applications. At billion-scale, however, existing graph embedding systems face a trade-off: they either rely on large in-memory footprints across many GPUs (limited scalability) or repeatedly stream data from disk (incurring severe I/O overhead and low GPU utilization). In this paper, we propose Legend, a lightweight hete- rogeneous system for graph embedding that systematically redesigns data management across CPU, GPU, and NVMe SSD resources. Legend combines three practical ideas: (1) a prefetch-friendly embedding-loading order that lets GPUs efficiently prefetch necessary embeddings directly from NVMe SSD with low I/O amplification; (2) a high-throughput GPU–SSD direct-access driver tuned for the access patterns of embedding training; and (3) a customized parallel execution strategy that maximizes GPU utilization. Together, these components let Legend store and stream vast embedding data without overprovisioning GPU memory or suffering I/O stalls. Extensive experiments on billion-scale graphs demonstrate that Legend speeds up end-to-end workloads by up to 4.8×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document} versus state-of-the-art systems, and matches their performance on the largest workloads while using only one quarter of the GPUs.
Zitieren
APA
Li, Z., Ke, X., Zhu, Y., Gao, Y., & Li, F. (2026). Efficient graph embedding at scale: optimizing CPU-GPU-SSD integration. The VLDB Journal, 35(3). https://doi.org/10.1007/s00778-026-00974-8
Vancouver
Li Z, Ke X, Zhu Y, Gao Y, Li F. Efficient graph embedding at scale: optimizing CPU-GPU-SSD integration. The VLDB Journal. 2026;35(3). doi:10.1007/s00778-026-00974-8
BibTeX
@article{Li2026,
author = {Li, Zhonggen and Ke, Xiangyu and Zhu, Yifan and Gao, Yunjun and Li, Feifei},
title = {Efficient graph embedding at scale: optimizing CPU-GPU-SSD integration},
journal = {The VLDB Journal},
year = {2026},
volume = {35},
number = {3},
doi = {10.1007/s00778-026-00974-8}
}
EUVIMED