Which can Accelerate Distributed Machine Learning Faster: Hybrid Optical/Electrical or Optical Reconfigurable DCN?

We run various distributed machine learning (DML) architectures in a hybrid optical/electrical DCN and an optical DCN based on Hyper-FleX-LION. Experimental results show that Hyper-FleX-LION gains faster DML acceleration and improves acceleration ratio by up to 22.3%.

Bibliographic Details
Published in:Optical Fiber Communication Conference (OFC) 2022
Main Authors: Yang H., Zhu Z., Proietti R., Ben Yoo S. J.
Other Authors: Yang, H., Zhu, Z., Proietti, R., Ben Yoo, S. J.
Format: Conference Object
Language:English
Published: Institute of Electrical and Electronics Engineers Inc. 2022
Subjects:
Distributed machine learning
optical switching
datacenter network
DML
Online Access:https://hdl.handle.net/11583/2973049
https://doi.org/10.1364/OFC.2022.Th1G.5
Description
Summary:We run various distributed machine learning (DML) architectures in a hybrid optical/electrical DCN and an optical DCN based on Hyper-FleX-LION. Experimental results show that Hyper-FleX-LION gains faster DML acceleration and improves acceleration ratio by up to 22.3%.

Similar Items