| Item type |
SIG Technical Reports(1) |
| 公開日 |
2018-09-20 |
| タイトル |
|
|
タイトル |
A Node Level Performance/Power Efficiency Aware Resource Management Technique |
| タイトル |
|
|
言語 |
en |
|
タイトル |
A Node Level Performance/Power Efficiency Aware Resource Management Technique |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
システム構築と運用 |
| 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 |
|
|
|
Graduate School of Informatics, Kyoto University |
| 著者所属 |
|
|
|
Academic Center for Computing and Media Studies, Kyoto University |
| 著者所属 |
|
|
|
Academic Center for Computing and Media Studies, Kyoto University |
| 著者所属 |
|
|
|
Research Institute for Information Technology Kyushu University |
| 著者所属(英) |
|
|
|
en |
|
|
Graduate School of Informatics, Kyoto University |
| 著者所属(英) |
|
|
|
en |
|
|
Academic Center for Computing and Media Studies, Kyoto University |
| 著者所属(英) |
|
|
|
en |
|
|
Academic Center for Computing and Media Studies, Kyoto University |
| 著者所属(英) |
|
|
|
en |
|
|
Research Institute for Information Technology Kyushu University |
| 著者名 |
Le, Li
Keiichiro, Fukazawa
Hiroshi, Nakashima
Takeshi, Nanri
|
| 著者名(英) |
Le, Li
Keiichiro, Fukazawa
Hiroshi, Nakashima
Takeshi, Nanri
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
The amount of data grows rapidly, since everything can be easily digitalized today. For obtaining values from those data, datacenters continuously enlarge and enhance their computer systems. However, recently the silicon transistors are difficult to develop, and it becomes much harder for us to extract its physical ultimate, and so the performance becomes hardly catch up the growth of data. Thus, we have to increase the number of hardware to reach the demanded performance and also sacrifice more power as a result. Most of the supercomputer systems are homogeneous systems, which means all of their nodes are comprised by identical hardware. However, the process variation in manufacturing processes results in a variation on performance/power. In this research, taking a node level performance power variation into consideration, we come up with a power saving resource management technique. We also make a job scheduling simulator to verify its effects by comparing with primary methods. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
The amount of data grows rapidly, since everything can be easily digitalized today. For obtaining values from those data, datacenters continuously enlarge and enhance their computer systems. However, recently the silicon transistors are difficult to develop, and it becomes much harder for us to extract its physical ultimate, and so the performance becomes hardly catch up the growth of data. Thus, we have to increase the number of hardware to reach the demanded performance and also sacrifice more power as a result. Most of the supercomputer systems are homogeneous systems, which means all of their nodes are comprised by identical hardware. However, the process variation in manufacturing processes results in a variation on performance/power. In this research, taking a node level performance power variation into consideration, we come up with a power saving resource management technique. We also make a job scheduling simulator to verify its effects by comparing with primary methods. |
| 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 2018-HPC-166,
号 3,
p. 1-7,
発行日 2018-09-20
|
| 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-HPC18166003.pdf (1.2 MB)
