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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Graph Neural Networks (GNNs) have demonstrated outstanding performance invarious applications. Existing frameworks utilize CPU-GPU heterogeneousenvironments to train GNN models and integrate mini-batch and samplingtechniques to overcome the GPU memory limitation. In CPU-GPU heterogeneousenvironments, we can divide sample-based GNN training into three steps: sample,gather, and train. Existing GNN systems use different task orchestratingmethods to employ each step on CPU or GPU. After extensive experiments andanalysis, we find that existing task orchestrating methods fail to fullyutilize the heterogeneous resources, limited by inefficient CPU processing orGPU resource contention. In this paper, we propose NeutronOrch, a system forsample-based GNN training that incorporates a layer-based task orchestratingmethod and ensures balanced utilization of the CPU and GPU. NeutronOrchdecouples the training process by layer and pushes down the training task ofthe bottom layer to the CPU. This significantly reduces the computational loadand memory footprint of GPU training. To avoid inefficient CPU processing,NeutronOrch only offloads the training of frequently accessed vertices to theCPU and lets GPU reuse their embeddings with bounded staleness. Furthermore,NeutronOrch provides a fine-grained pipeline design for the layer-based taskorchestrating method, fully overlapping different tasks on heterogeneousresources while strictly guaranteeing bounded staleness. The experimentalresults show that compared with the state-of-the-art GNN systems, NeutronOrchcan achieve up to 11.51x performance speedup.

NeutronOrch: Rethinking Sample-based GNN Training under CPU-GPU Heterogeneous Environments