| Item type |
SIG Technical Reports(1) |
| 公開日 |
2015-07-28 |
| タイトル |
|
|
タイトル |
Analyzing Task Parallel Program Traces based on DAG Visualization |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Analyzing Task Parallel Program Traces based on DAG Visualization |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
性能評価ツール・モデル |
| 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 |
|
|
|
University of Tokyo |
| 著者所属 |
|
|
|
University of Notre Dame |
| 著者所属 |
|
|
|
Chalmers University of Technolog |
| 著者所属 |
|
|
|
University of Tokyo |
| 著者所属(英) |
|
|
|
en |
|
|
University of Tokyo |
| 著者所属(英) |
|
|
|
en |
|
|
University of Notre Dame |
| 著者所属(英) |
|
|
|
en |
|
|
Chalmers University of Technolog |
| 著者所属(英) |
|
|
|
en |
|
|
University of Tokyo |
| 著者名 |
An, Huynh
Douglas, Thain
Miquel, Pericas
Kenjiro, Taura
|
| 著者名(英) |
An, Huynh
Douglas, Thain
Miquel, Pericas
Kenjiro, Taura
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Task parallel programming models are conceived as a promising paradigm that brings intricate parallel programming techniques to a larger audience of programmers because of its high programmability. Programmers just need to expose parallelism in their programs by creating “task”, which is a lightweight work unit that can execute in parallel with the rest, at arbitrary places in their code, then all other low-level burdens in the parallel execution of these tasks such as thread management, task scheduling, and load balancing are handled automatically by runtime systems. However, this dependence on runtime systems also hides execution mechanisms of a task parallel program from programmers, making it difficult for them to understand the cause of suboptimal performance of their programs. As an effort to tackle this problem, we have developed tools that capture and visualize the trace of an execution of a task parallel program in the form of a directed acyclic graph (DAG) which is augmented with relevant performance information on the execution. Each node of the DAG is an interval of a serial code segment in the program. Each edge represents a dependency between two nodes. This DAG visualization provides a task-centric view of the program, which is different from other popular visualizations such as thread-centric timeline visualization and code-centric hotspots analysis. Our tool (DAGViz) displays a DAG in a hierarchical manner, expands to more detailed views on demand which helps users view aggregate performance information at various levels of detail. Beside DAG view, DAGViz also provides a timeline visualization which is coordinated with the DAG. This coordination helps users relate interesting points between thread-centric timeline and task-centric DAG structure. DAGViz is expected to support effectively the process of analyzing task parallel performance and developing scheduling algorithms for task parallel schedulers. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Task parallel programming models are conceived as a promising paradigm that brings intricate parallel programming techniques to a larger audience of programmers because of its high programmability. Programmers just need to expose parallelism in their programs by creating “task”, which is a lightweight work unit that can execute in parallel with the rest, at arbitrary places in their code, then all other low-level burdens in the parallel execution of these tasks such as thread management, task scheduling, and load balancing are handled automatically by runtime systems. However, this dependence on runtime systems also hides execution mechanisms of a task parallel program from programmers, making it difficult for them to understand the cause of suboptimal performance of their programs. As an effort to tackle this problem, we have developed tools that capture and visualize the trace of an execution of a task parallel program in the form of a directed acyclic graph (DAG) which is augmented with relevant performance information on the execution. Each node of the DAG is an interval of a serial code segment in the program. Each edge represents a dependency between two nodes. This DAG visualization provides a task-centric view of the program, which is different from other popular visualizations such as thread-centric timeline visualization and code-centric hotspots analysis. Our tool (DAGViz) displays a DAG in a hierarchical manner, expands to more detailed views on demand which helps users view aggregate performance information at various levels of detail. Beside DAG view, DAGViz also provides a timeline visualization which is coordinated with the DAG. This coordination helps users relate interesting points between thread-centric timeline and task-centric DAG structure. DAGViz is expected to support effectively the process of analyzing task parallel performance and developing scheduling algorithms for task parallel schedulers. |
| 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 2015-HPC-150,
号 34,
p. 1-9,
発行日 2015-07-28
|
| 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-HPC15150034.pdf (1.8 MB)
