Enregistré dans:
Détails bibliographiques
Auteur principal: Jeaugey, Sylvain
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2506.20252
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866915358337662976
author Jeaugey, Sylvain
author_facet Jeaugey, Sylvain
contents This paper describes a new algorithm called PAT, for Parallel Aggregated Trees, and which can be used to implement all-gather and reduce-scatter operations. This algorithm works on any number of ranks, has a logarithmic number of network transfers for small size operations, minimizes long-distance communication, and requires a logarithmic amount of internal buffers, independently from the total operation size. It is aimed at improving the performance of the NCCL library in cases where the ring algorithm would be inefficient, as its linear latency would show poor performance for small sizes and/or at scale.
format Preprint
id arxiv_https___arxiv_org_abs_2506_20252
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PAT: a new algorithm for all-gather and reduce-scatter operations at scale
Jeaugey, Sylvain
Distributed, Parallel, and Cluster Computing
This paper describes a new algorithm called PAT, for Parallel Aggregated Trees, and which can be used to implement all-gather and reduce-scatter operations. This algorithm works on any number of ranks, has a logarithmic number of network transfers for small size operations, minimizes long-distance communication, and requires a logarithmic amount of internal buffers, independently from the total operation size. It is aimed at improving the performance of the NCCL library in cases where the ring algorithm would be inefficient, as its linear latency would show poor performance for small sizes and/or at scale.
title PAT: a new algorithm for all-gather and reduce-scatter operations at scale
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2506.20252