| Item type |
SIG Technical Reports(1) |
| 公開日 |
2017-09-12 |
| タイトル |
|
|
タイトル |
A Performance Evaluation of Distributed TensorFlow |
| タイトル |
|
|
言語 |
en |
|
タイトル |
A Performance Evaluation of Distributed TensorFlow |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
分散ファイルシステムと機械学習 |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
筑波大学／産業技術総合研究所 |
| 著者所属 |
|
|
|
産業技術総合研究所／筑波大学 |
| 著者所属 |
|
|
|
産業技術総合研究所 |
| 著者所属 |
|
|
|
産業技術総合研究所／筑波大学 |
| 著者所属(英) |
|
|
|
en |
|
|
University of Tsukuba / National Institute of Advanced Industrial Science and Technology |
| 著者所属(英) |
|
|
|
en |
|
|
National Institute of Advanced Industrial Science and Technology / University of Tsukuba |
| 著者所属(英) |
|
|
|
en |
|
|
National Institute of Advanced Industrial Science and Technology |
| 著者所属(英) |
|
|
|
en |
|
|
National Institute of Advanced Industrial Science and Technology / University of Tsukuba |
| 著者名 |
Tianlun, Wang
Yusuke, Tanimura
Hirotaka, Ogawa
Hidemoto, Nakada
|
| 著者名(英) |
Tianlun, Wang
Yusuke, Tanimura
Hirotaka, Ogawa
Hidemoto, Nakada
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
TensorFlow is a deep learning framework which is developed by Google. The computations in TensorFlow are implemented and expressed as data flow graphs of multidimensional array data which is referred to as“Tensor.” We know that, with TensorFlow we can implement parallel execution in a multi-GPU configuration and distributed execution in a multi-node configuration, but it is not clear how effective it is in the real environment. In this paper, we measured these performances for several mini batch sizes and network settings. From the experimental results, we confirmed that we can accelerate the execution in all the environment we have tested. We also found that the mini batch size has a big influence in the distributed environment of 1Gbps network. However, in the environment of 10 Gbps network or the intra-node multipl-GPU configuration, we could achieve the linear speed up regardless of mini batch size. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
TensorFlow is a deep learning framework which is developed by Google. The computations in TensorFlow are implemented and expressed as data flow graphs of multidimensional array data which is referred to as“Tensor.” We know that, with TensorFlow we can implement parallel execution in a multi-GPU configuration and distributed execution in a multi-node configuration, but it is not clear how effective it is in the real environment. In this paper, we measured these performances for several mini batch sizes and network settings. From the experimental results, we confirmed that we can accelerate the execution in all the environment we have tested. We also found that the mini batch size has a big influence in the distributed environment of 1Gbps network. However, in the environment of 10 Gbps network or the intra-node multipl-GPU configuration, we could achieve the linear speed up regardless of mini batch size. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AN10463942 |
| 書誌情報 |
研究報告ハイパフォーマンスコンピューティング（HPC）
巻 2017-HPC-161,
号 1,
p. 1-6,
発行日 2017-09-12
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-8841 |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-HPC17161001.pdf (2.7 MB)
