MixNet: A Runtime Reconfigurable Optical-Electrical Fabric for Distributed Mixture-of-Experts Training

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Liao, Xudong, Sun, Yijun, Tian, Han, Wan, Xinchen, Jin, Yilun, Wang, Zilong, Ren, Zhenghang, Huang, Xinyang, Li, Wenxue, Tse, Kin Fai, Zhong, Zhizhen, Liu, Guyue, Zhang, Ying, Ye, Xiaofeng, Zhang, Yiming, Chen, Kai
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909769545023488
author Liao, Xudong
Sun, Yijun
Tian, Han
Wan, Xinchen
Jin, Yilun
Wang, Zilong
Ren, Zhenghang
Huang, Xinyang
Li, Wenxue
Tse, Kin Fai
Zhong, Zhizhen
Liu, Guyue
Zhang, Ying
Ye, Xiaofeng
Zhang, Yiming
Chen, Kai
author_facet Liao, Xudong
Sun, Yijun
Tian, Han
Wan, Xinchen
Jin, Yilun
Wang, Zilong
Ren, Zhenghang
Huang, Xinyang
Li, Wenxue
Tse, Kin Fai
Zhong, Zhizhen
Liu, Guyue
Zhang, Ying
Ye, Xiaofeng
Zhang, Yiming
Chen, Kai
contents Mixture-of-Expert (MoE) models outperform conventional models by selectively activating different subnets, named experts, on a per-token basis. This gated computation generates dynamic communications that cannot be determined beforehand, challenging the existing GPU interconnects that remain static during the distributed training process. In this paper, we advocate for a first-of-its-kind system, called MixNet, that unlocks topology reconfiguration during distributed MoE training. Towards this vision, we first perform a production measurement study and show that the MoE dynamic communication pattern has strong locality, alleviating the requirement of global reconfiguration. Based on this, we design and implement a regionally reconfigurable high-bandwidth domain on top of existing electrical interconnects using optical circuit switching (OCS), achieving scalability while maintaining rapid adaptability. We have built a fully functional MixNet prototype with commodity hardware and a customized collective communication runtime that trains state-of-the-art MoE models with in-training topology reconfiguration across 32 A100 GPUs. Large-scale packet-level simulations show that MixNet delivers comparable performance as the non-blocking fat-tree fabric while boosting the training cost efficiency (e.g., performance per dollar) of four representative MoE models by 1.2x-1.5x and 1.9x-2.3x at 100 Gbps and 400 Gbps link bandwidths, respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2501_03905
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MixNet: A Runtime Reconfigurable Optical-Electrical Fabric for Distributed Mixture-of-Experts Training
Liao, Xudong
Sun, Yijun
Tian, Han
Wan, Xinchen
Jin, Yilun
Wang, Zilong
Ren, Zhenghang
Huang, Xinyang
Li, Wenxue
Tse, Kin Fai
Zhong, Zhizhen
Liu, Guyue
Zhang, Ying
Ye, Xiaofeng
Zhang, Yiming
Chen, Kai
Networking and Internet Architecture
Machine Learning
Mixture-of-Expert (MoE) models outperform conventional models by selectively activating different subnets, named experts, on a per-token basis. This gated computation generates dynamic communications that cannot be determined beforehand, challenging the existing GPU interconnects that remain static during the distributed training process. In this paper, we advocate for a first-of-its-kind system, called MixNet, that unlocks topology reconfiguration during distributed MoE training. Towards this vision, we first perform a production measurement study and show that the MoE dynamic communication pattern has strong locality, alleviating the requirement of global reconfiguration. Based on this, we design and implement a regionally reconfigurable high-bandwidth domain on top of existing electrical interconnects using optical circuit switching (OCS), achieving scalability while maintaining rapid adaptability. We have built a fully functional MixNet prototype with commodity hardware and a customized collective communication runtime that trains state-of-the-art MoE models with in-training topology reconfiguration across 32 A100 GPUs. Large-scale packet-level simulations show that MixNet delivers comparable performance as the non-blocking fat-tree fabric while boosting the training cost efficiency (e.g., performance per dollar) of four representative MoE models by 1.2x-1.5x and 1.9x-2.3x at 100 Gbps and 400 Gbps link bandwidths, respectively.
title MixNet: A Runtime Reconfigurable Optical-Electrical Fabric for Distributed Mixture-of-Experts Training
topic Networking and Internet Architecture
Machine Learning
url https://arxiv.org/abs/2501.03905