gZCCL: Compression-Accelerated Collective Communication Framework for GPU Clusters

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Huang, Jiajun, Di, Sheng, Yu, Xiaodong, Zhai, Yujia, Liu, Jinyang, Huang, Yafan, Raffenetti, Ken, Zhou, Hui, Zhao, Kai, Lu, Xiaoyi, Chen, Zizhong, Cappello, Franck, Guo, Yanfei, Thakur, Rajeev
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917658823229440
author Huang, Jiajun
Di, Sheng
Yu, Xiaodong
Zhai, Yujia
Liu, Jinyang
Huang, Yafan
Raffenetti, Ken
Zhou, Hui
Zhao, Kai
Lu, Xiaoyi
Chen, Zizhong
Cappello, Franck
Guo, Yanfei
Thakur, Rajeev
author_facet Huang, Jiajun
Di, Sheng
Yu, Xiaodong
Zhai, Yujia
Liu, Jinyang
Huang, Yafan
Raffenetti, Ken
Zhou, Hui
Zhao, Kai
Lu, Xiaoyi
Chen, Zizhong
Cappello, Franck
Guo, Yanfei
Thakur, Rajeev
contents GPU-aware collective communication has become a major bottleneck for modern computing platforms as GPU computing power rapidly rises. A traditional approach is to directly integrate lossy compression into GPU-aware collectives, which can lead to serious performance issues such as underutilized GPU devices and uncontrolled data distortion. In order to address these issues, in this paper, we propose gZCCL, a first-ever general framework that designs and optimizes GPU-aware, compression-enabled collectives with an accuracy-aware design to control error propagation. To validate our framework, we evaluate the performance on up to 512 NVIDIA A100 GPUs with real-world applications and datasets. Experimental results demonstrate that our gZCCL-accelerated collectives, including both collective computation (Allreduce) and collective data movement (Scatter), can outperform NCCL as well as Cray MPI by up to 4.5X and 28.7X, respectively. Furthermore, our accuracy evaluation with an image-stacking application confirms the high reconstructed data quality of our accuracy-aware framework.
format Preprint
id arxiv_https___arxiv_org_abs_2308_05199
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle gZCCL: Compression-Accelerated Collective Communication Framework for GPU Clusters
Huang, Jiajun
Di, Sheng
Yu, Xiaodong
Zhai, Yujia
Liu, Jinyang
Huang, Yafan
Raffenetti, Ken
Zhou, Hui
Zhao, Kai
Lu, Xiaoyi
Chen, Zizhong
Cappello, Franck
Guo, Yanfei
Thakur, Rajeev
Distributed, Parallel, and Cluster Computing
GPU-aware collective communication has become a major bottleneck for modern computing platforms as GPU computing power rapidly rises. A traditional approach is to directly integrate lossy compression into GPU-aware collectives, which can lead to serious performance issues such as underutilized GPU devices and uncontrolled data distortion. In order to address these issues, in this paper, we propose gZCCL, a first-ever general framework that designs and optimizes GPU-aware, compression-enabled collectives with an accuracy-aware design to control error propagation. To validate our framework, we evaluate the performance on up to 512 NVIDIA A100 GPUs with real-world applications and datasets. Experimental results demonstrate that our gZCCL-accelerated collectives, including both collective computation (Allreduce) and collective data movement (Scatter), can outperform NCCL as well as Cray MPI by up to 4.5X and 28.7X, respectively. Furthermore, our accuracy evaluation with an image-stacking application confirms the high reconstructed data quality of our accuracy-aware framework.
title gZCCL: Compression-Accelerated Collective Communication Framework for GPU Clusters
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2308.05199