Guardado en:
Detalles Bibliográficos
Autores principales: Xu, Chao, Zhang, Xu, Luo, Zihang, Wu, Yuyan, Qian, Guoxin, Yao, Yufeng, Wang, Chihyung, Zhou, Jingbin
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2605.22428
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911704641699840
author Xu, Chao
Zhang, Xu
Luo, Zihang
Wu, Yuyan
Qian, Guoxin
Yao, Yufeng
Wang, Chihyung
Zhou, Jingbin
author_facet Xu, Chao
Zhang, Xu
Luo, Zihang
Wu, Yuyan
Qian, Guoxin
Yao, Yufeng
Wang, Chihyung
Zhou, Jingbin
contents Reducing collective communication latency is a critical goal for large model training and inference in both academia and industry. Many-to-many communications, such as AllGather and AlltoAll (dispatch), are core components of modern parallelization strategies. State-of-the-art implementations of these communications rely on unicast-based writes and transmit duplicate copies of the same data across physical links for multiple receivers. This redundant transmission congests network bottlenecks and degrades end-to-end latency. We present MultiWrite, a novel many-to-many transmission semantic that eliminates redundant packets to directly reduce operator latency. MultiWrite adopts multicast principles while addressing critical limitations of traditional multicast for AI workloads. These limitations include heavy management plane overhead and ecosystem compatibility issues. We implement MultiWrite on Ascend NPUs. Long-term stress tests demonstrate that our MultiWrite-based operators achieve up to 33% latency reduction on commercially deployed devices.
format Preprint
id arxiv_https___arxiv_org_abs_2605_22428
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Exploiting Multicast for Accelerating Collective Communication
Xu, Chao
Zhang, Xu
Luo, Zihang
Wu, Yuyan
Qian, Guoxin
Yao, Yufeng
Wang, Chihyung
Zhou, Jingbin
Distributed, Parallel, and Cluster Computing
Reducing collective communication latency is a critical goal for large model training and inference in both academia and industry. Many-to-many communications, such as AllGather and AlltoAll (dispatch), are core components of modern parallelization strategies. State-of-the-art implementations of these communications rely on unicast-based writes and transmit duplicate copies of the same data across physical links for multiple receivers. This redundant transmission congests network bottlenecks and degrades end-to-end latency. We present MultiWrite, a novel many-to-many transmission semantic that eliminates redundant packets to directly reduce operator latency. MultiWrite adopts multicast principles while addressing critical limitations of traditional multicast for AI workloads. These limitations include heavy management plane overhead and ecosystem compatibility issues. We implement MultiWrite on Ascend NPUs. Long-term stress tests demonstrate that our MultiWrite-based operators achieve up to 33% latency reduction on commercially deployed devices.
title Exploiting Multicast for Accelerating Collective Communication
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2605.22428