CBA: Communication-Bound-Aware Cross-Domain Resource Assignment for Pipeline-Parallel Distributed LLM Training in Dynamic Multi-DC Optical Networks

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Fu, Dianxuan, Liu, Xiaomin, Zhang, Yihao, Shen, Shikui, Hu, Weisheng, Zhuge, Qunbi
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915692316459008
author Fu, Dianxuan
Liu, Xiaomin
Zhang, Yihao
Shen, Shikui
Hu, Weisheng
Zhuge, Qunbi
author_facet Fu, Dianxuan
Liu, Xiaomin
Zhang, Yihao
Shen, Shikui
Hu, Weisheng
Zhuge, Qunbi
contents We propose a communication-bound-aware cross-domain resource assignment framework for pipeline-parallel distributed training over multi-datacenter optical networks, which lowers iteration time by 31.25% and reduces 13.20% blocking requests compared to baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2512_20080
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CBA: Communication-Bound-Aware Cross-Domain Resource Assignment for Pipeline-Parallel Distributed LLM Training in Dynamic Multi-DC Optical Networks
Fu, Dianxuan
Liu, Xiaomin
Zhang, Yihao
Shen, Shikui
Hu, Weisheng
Zhuge, Qunbi
Networking and Internet Architecture
Artificial Intelligence
Optics
We propose a communication-bound-aware cross-domain resource assignment framework for pipeline-parallel distributed training over multi-datacenter optical networks, which lowers iteration time by 31.25% and reduces 13.20% blocking requests compared to baselines.
title CBA: Communication-Bound-Aware Cross-Domain Resource Assignment for Pipeline-Parallel Distributed LLM Training in Dynamic Multi-DC Optical Networks
topic Networking and Internet Architecture
Artificial Intelligence
Optics
url https://arxiv.org/abs/2512.20080