Hybrid Electrical/Optical Switch Architectures for Training Distributed Deep Learning in Large-Scale

Thao-Nguyen TRUONG  Ryousei TAKANO  

IEICE TRANSACTIONS on Information and Systems   Vol.E104-D   No.8   pp.1332-1339
Publication Date: 2021/08/01
Publicized: 2021/04/23
Online ISSN: 1745-1361
DOI: 10.1587/transinf.2020EDP7201
Type of Manuscript: PAPER
Category: Information Network
distributed deep learning,  high performance computing (HPC),  optical circuit switching,  hybrid switching,  

Full Text: PDF(1MB)>>
Buy this Article

Data parallelism is the dominant method used to train deep learning (DL) models on High-Performance Computing systems such as large-scale GPU clusters. When training a DL model on a large number of nodes, inter-node communication becomes bottle-neck due to its relatively higher latency and lower link bandwidth (than intra-node communication). Although some communication techniques have been proposed to cope with this problem, all of these approaches target to deal with the large message size issue while diminishing the effect of the limitation of the inter-node network. In this study, we investigate the benefit of increasing inter-node link bandwidth by using hybrid switching systems, i.e., Electrical Packet Switching and Optical Circuit Switching. We found that the typical data-transfer of synchronous data-parallelism training is long-lived and rarely changed that can be speed-up with optical switching. Simulation results on the Simgrid simulator show that our approach speed-up the training time of deep learning applications, especially in a large-scale manner.