| Item type |
SIG Technical Reports(1) |
| 公開日 |
2016-08-01 |
| タイトル |
|
|
タイトル |
Performance Modeling of Task Parallel Programs |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Performance Modeling of Task Parallel Programs |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
並列処理 |
| 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 |
|
|
|
The University of Tokyo |
| 著者所属 |
|
|
|
The University of Tokyo |
| 著者所属(英) |
|
|
|
en |
|
|
The University of Tokyo |
| 著者所属(英) |
|
|
|
en |
|
|
The University of Tokyo |
| 著者名 |
Namsraijav, Byambajav
Kenjiro, Taura
|
| 著者名(英) |
Namsraijav, Byambajav
Kenjiro, Taura
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Task parallel programming makes it easy for programmers to write parallel applications by removing the burden of dealing with low-level details of thread management, task scheduling, and load balancing. Since task parallel run-time systems employ dynamic work-stealing scheduler for running an application on multiple threads, the performance modeling of a task parallel program, i.e. how the program performs as the number of cores increases or how long it executes on a different input, is hard to predict. This paper proposes a method which combines profiling based analytical performance modeling techniques with regression-based models. Our evaluation shows that predicting the execution time of task parallel applications using the proposed two-step method is significantly more accurate than predicting the execution time directly with regression models. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Task parallel programming makes it easy for programmers to write parallel applications by removing the burden of dealing with low-level details of thread management, task scheduling, and load balancing. Since task parallel run-time systems employ dynamic work-stealing scheduler for running an application on multiple threads, the performance modeling of a task parallel program, i.e. how the program performs as the number of cores increases or how long it executes on a different input, is hard to predict. This paper proposes a method which combines profiling based analytical performance modeling techniques with regression-based models. Our evaluation shows that predicting the execution time of task parallel applications using the proposed two-step method is significantly more accurate than predicting the execution time directly with regression models. |
| 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 2016-HPC-155,
号 24,
p. 1-8,
発行日 2016-08-01
|
| 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-HPC16155024.pdf (571.4 kB)
