A distributed synchronous SGD algorithm with global top-k Sparsification for low bandwidth networks

Shaohuai Shi, Qiang WANG, Kaiyong Zhao, Zhenheng Tang, Yuxin Wang, Xiang Huang, Xiaowen CHU*

*Corresponding author for this work

Research output: Chapter in book/report/conference proceedingConference contributionpeer-review

7 Citations (Scopus)

Abstract

Distributed synchronous stochastic gradient descent (S-SGD) with data parallelism has been widely used in training large-scale deep neural networks (DNNs), but it typically requires very high communication bandwidth between computational workers (e.g., GPUs) to exchange gradients iteratively. Recently, Top-k sparsification techniques have been proposed to reduce the volume of data to be exchanged among workers and thus alleviate the network pressure. Top-k sparsification can zero-out a significant portion of gradients without impacting the model convergence. However, the sparse gradients should be transferred with their indices, and the irregular indices make the sparse gradients aggregation difficult. Current methods that use AllGather to accumulate the sparse gradients have a communication complexity of O(kP), where P is the number of workers, which is inefficient on low bandwidth networks with a large number of workers. We observe that not all top-k gradients from P workers are needed for the model update, and therefore we propose a novel global Top-k (gTop-k) sparsification mechanism to address the difficulty of aggregating sparse gradients. Specifically, we choose global top-k largest absolute values of gradients from P workers, instead of accumulating all local top-k gradients to update the model in each iteration. The gradient aggregation method based on gTop-k sparsification, namely gTopKAllReduce, reduces the communication complexity from O(kP) to O(k log P). Through extensive experiments on different DNNs, we verify that gTop-k S-SGD has nearly consistent convergence performance with S-SGD, and it has only slight degradations on generalization performance. In terms of scaling efficiency, we evaluate gTop-k on a cluster with 32 GPU machines which are interconnected with 1 Gbps Ethernet. The experimental results show that our method achieves 2.7-12× higher scaling efficiency than S-SGD with dense gradients and 1.1-1.7× improvement than the existing Top-k S-SGD.

Original languageEnglish
Title of host publicationProceedings - 2019 39th IEEE International Conference on Distributed Computing Systems, ICDCS 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages2238-2247
Number of pages10
ISBN (Electronic)9781728125190
DOIs
Publication statusPublished - Jul 2019
Event39th IEEE International Conference on Distributed Computing Systems, ICDCS 2019 - Richardson, United States
Duration: 7 Jul 20199 Jul 2019

Publication series

NameProceedings - International Conference on Distributed Computing Systems
Volume2019-July

Conference

Conference39th IEEE International Conference on Distributed Computing Systems, ICDCS 2019
Country/TerritoryUnited States
CityRichardson
Period7/07/199/07/19

Scopus Subject Areas

  • Software
  • Hardware and Architecture
  • Computer Networks and Communications

User-Defined Keywords

  • Deep Learning
  • Distributed SGD
  • Gradient Communication
  • gTop-k
  • Stochastic Gradient Descent
  • Top-k Sparsification

Fingerprint

Dive into the research topics of 'A distributed synchronous SGD algorithm with global top-k Sparsification for low bandwidth networks'. Together they form a unique fingerprint.

Cite this