I/O Characterization and Performance Evaluation of BeeGFS for Deep Learning

I/O Characterization and Performance Evaluation of BeeGFS for Deep Learning
复制标题

BeeGFS 深度学习的 I/O 表征和性能评估

DOI:
10.1145/3337821.3337902
复制
发表时间:
2019
期刊:
ICPP 2019: Proceedings of the 48th International Conference on Parallel Processing
影响因子:
--
通讯作者:
Yu, Weikuan
Yu, Weikuan
中科院分区:
--
文献类型:
--
作者:
Chowdhury, Fahim;Zhu, Yue;Heer, Todd;Paredes, Saul;Moody, Adam;Goldstone, Robin;Mohror, Kathryn;Yu, Weikuan

文献摘要

参考文献

被引文献

相似文献

并行文件系统 (PFS) 经常部署在领先的高性能计算 (HPC) 系统上,以确保高效的 I/O、持久存储和可扩展性能。新兴的深度学习 (DL) 应用程序对批量输入小型随机文件的 HPC 系统提出了新的 I/O 和存储要求。这要求 PFS 具有能够满足 DL 应用程序需求的相应功能。 BeeGFS 是最近兴起的 PFS,因其性能、可扩展性和易用性而引起了研究界和工业界的关注。在强调对 BeeGFS 进行系统性能分析的同时,本文介绍了 BeeGFS 的架构和系统特征,并使用尖端的 I/O、元数据和深度学习应用基准进行实验评估。特别是,我们使用了 AlexNet 和 ResNet-50 模型,使用 Livermore Big Artificial Neural Network Toolkit (LBANN) 和基于 TensorFlow 和 Horovod 的 ImageNet 数据读取器管道对 ImageNet 数据集进行分类。通过对 BeeGFS 的广泛性能表征,我们的研究提供了有关如何将 BeeGFS 用于新兴 DL 应用程序的有用文档。
Parallel File Systems (PFSs) are frequently deployed on leadership High Performance Computing (HPC) systems to ensure efficient I/O, persistent storage and scalable performance. Emerging Deep Learning (DL) applications incur new I/O and storage requirements to HPC systems with batched input of small random files. This mandates PFSs to have commensurate features that can meet the needs of DL applications. BeeGFS is a recently emerging PFS that has grabbed the attention of the research and industry world because of its performance, scalability and ease of use. While emphasizing a systematic performance analysis of BeeGFS, in this paper, we present the architectural and system features of BeeGFS, and perform an experimental evaluation using cutting-edge I/O, Metadata and DL application benchmarks. Particularly, we have utilized AlexNet and ResNet-50 models for the classification of ImageNet dataset using the Livermore Big Artificial Neural Network Toolkit (LBANN), and ImageNet data reader pipeline atop TensorFlow and Horovod. Through extensive performance characterization of BeeGFS, our study provides a useful documentation on how to leverage BeeGFS for the emerging DL applications.
集群文件系统高速互连的优势:Lustre 案例研究
DOI: 10.1109/ipdps.2006.1639564
发表时间: 2006
期刊: Proceedings 20th IEEE International Parallel & Distributed Processing Symposium
影响因子: --
作者:
Weikuan Yu;R. Noronha;Shuang Liang;D. Panda
通讯作者: D. Panda
DOI: 10.1109/mascots.2018.00023
发表时间: 2018-09
期刊: 2018 IEEE 26th International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems (MASCOTS)
影响因子: --
作者:
Yue Zhu;Fahim Chowdhury;Huansong Fu;A. Moody;K. Mohror;Kento Sato;Weikuan Yu
通讯作者: Yue Zhu;Fahim Chowdhury;Huansong Fu;A. Moody;K. Mohror;Kento Sato;Weikuan Yu
DOI: 10.1007/978-3-319-70581-1_17
发表时间: 2017-08
期刊: ArXiv
影响因子: --
作者:
Yuriy Kochura;S. Stirenko;Oleg Alienin;Michail Novotarskiy;Yuri G. Gordienko
通讯作者: Yuriy Kochura;S. Stirenko;Oleg Alienin;Michail Novotarskiy;Yuri G. Gordienko
DOI: --
发表时间: 2017
期刊: International Conference on Computer Science and Information Technologies
影响因子: --
作者:
Yuriy Kochura;S. Stirenko;A. Rojbi;Oleg Alienin;Michail Novotarskiy;Yuri G. Gordienko
通讯作者: Yuri G. Gordienko