Photonic Rails in ML Datacenters

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ding, Eric, Ouyang, Chuhan, Singh, Rachee
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911050176135168
author Ding, Eric
Ouyang, Chuhan
Singh, Rachee
author_facet Ding, Eric
Ouyang, Chuhan
Singh, Rachee
contents Rail-optimized network fabrics have become the de facto datacenter scale-out fabric for large-scale ML training. However, the use of high-radix electrical switches to provide all-to-all connectivity in rails imposes massive power, cost, and complexity overheads. We propose a rethinking of the rail abstraction by retaining its communication semantics, but realizing it using optical circuit switches. The key challenge is that optical switches support only one-to-one connectivity at a time, limiting the fan-out of traffic in ML workloads using hybrid parallelisms. We introduce parallelism-driven rail reconfiguration as a solution that leverages the sequential ordering between traffic from different parallelisms. We design a control plane, Opus, to enable time-multiplexed emulation of electrical rail switches using optical switches. More broadly, our work discusses a new research agenda: datacenter fabrics that co-evolve with the model parallelism dimensions within each job, as opposed to the prevailing mindset of reconfiguring networks before a job begins.
format Preprint
id arxiv_https___arxiv_org_abs_2507_08119
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Photonic Rails in ML Datacenters
Ding, Eric
Ouyang, Chuhan
Singh, Rachee
Networking and Internet Architecture
Rail-optimized network fabrics have become the de facto datacenter scale-out fabric for large-scale ML training. However, the use of high-radix electrical switches to provide all-to-all connectivity in rails imposes massive power, cost, and complexity overheads. We propose a rethinking of the rail abstraction by retaining its communication semantics, but realizing it using optical circuit switches. The key challenge is that optical switches support only one-to-one connectivity at a time, limiting the fan-out of traffic in ML workloads using hybrid parallelisms. We introduce parallelism-driven rail reconfiguration as a solution that leverages the sequential ordering between traffic from different parallelisms. We design a control plane, Opus, to enable time-multiplexed emulation of electrical rail switches using optical switches. More broadly, our work discusses a new research agenda: datacenter fabrics that co-evolve with the model parallelism dimensions within each job, as opposed to the prevailing mindset of reconfiguring networks before a job begins.
title Photonic Rails in ML Datacenters
topic Networking and Internet Architecture
url https://arxiv.org/abs/2507.08119