English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Graph Neural Networks (GNNs) have shown success in many real-worldapplications that involve graph-structured data. Most of the existingsingle-node GNN training systems are capable of training medium-scale graphswith tens of millions of edges; however, scaling them to large-scale graphswith billions of edges remains challenging. In addition, it is challenging tomap GNN training algorithms onto a computation node as state-of-the-artmachines feature heterogeneous architecture consisting of multiple processorsand a variety of accelerators.  We propose HyScale-GNN, a novel system to train GNN models on a single-nodeheterogeneous architecture. HyScale- GNN performs hybrid training whichutilizes both the processors and the accelerators to train a modelcollaboratively. Our system design overcomes the memory size limitation ofexisting works and is optimized for training GNNs on large-scale graphs. Wepropose a two-stage data pre-fetching scheme to reduce the communicationoverhead during GNN training. To improve task mapping efficiency, we propose adynamic resource management mechanism, which adjusts the workload assignmentand resource allocation during runtime. We evaluate HyScale-GNN on a CPU-GPUand a CPU-FPGA heterogeneous architecture. Using several large-scale datasetsand two widely-used GNN models, we compare the performance of our design with amulti-GPU baseline implemented in PyTorch-Geometric. The CPU-GPU design and theCPU-FPGA design achieve up to 2.08x speedup and 12.6x speedup, respectively.Compared with the state-of-the-art large-scale multi-node GNN training systemssuch as P3 and DistDGL, our CPU-FPGA design achieves up to 5.27x speedup usinga single node.

HyScale-GNN: A Scalable Hybrid GNN Training System on Single-Node Heterogeneous Architecture