Modeling the Impact of Fiber Latency on Compute-Communication Overlap in Geo-Distributed Multi-Datacenter AI Training

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Papavasileiou, Ioannis, Prabhakar, Sairam, Deo, Indu Kant, Makovejs, Sergejs
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913144192892928
author Papavasileiou, Ioannis
Prabhakar, Sairam
Deo, Indu Kant
Makovejs, Sergejs
author_facet Papavasileiou, Ioannis
Prabhakar, Sairam
Deo, Indu Kant
Makovejs, Sergejs
contents We use discrete-event simulation to quantify the impact of fiber latency on the efficacy of geo-distributed AI model training with data parallelism. We conclude that the optimum distances between two AI clusters is 10-100km, over which hollow-core fiber enables 25% higher compute-communication overlap.
format Preprint
id arxiv_https___arxiv_org_abs_2605_19169
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Modeling the Impact of Fiber Latency on Compute-Communication Overlap in Geo-Distributed Multi-Datacenter AI Training
Papavasileiou, Ioannis
Prabhakar, Sairam
Deo, Indu Kant
Makovejs, Sergejs
Performance
Distributed, Parallel, and Cluster Computing
We use discrete-event simulation to quantify the impact of fiber latency on the efficacy of geo-distributed AI model training with data parallelism. We conclude that the optimum distances between two AI clusters is 10-100km, over which hollow-core fiber enables 25% higher compute-communication overlap.
title Modeling the Impact of Fiber Latency on Compute-Communication Overlap in Geo-Distributed Multi-Datacenter AI Training
topic Performance
Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2605.19169