Exploiting Multicast for Accelerating Collective Communication
摘要
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.
关键词
引用
@article{arxiv.2605.22428,
title = {Exploiting Multicast for Accelerating Collective Communication},
author = {Chao Xu and Xu Zhang and Zihang Luo and Yuyan Wu and Guoxin Qian and Yufeng Yao and Chihyung Wang and Jingbin Zhou},
journal= {arXiv preprint arXiv:2605.22428},
year = {2026}
}