Visualization and analysis of parallel dataflow execution with smart traces

Daniel K. Osmari, Huy T. Vo, Claudio T. Silva, João L D Comba, Lauro Lins

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Most performance analysis tools focus on presenting an overload of details, with little application-dependent structure, and predefined statistical summaries. This makes the complex relations present in a parallel program not directly recognisable to the user, making the task of identifying performance issues more costly in both time and effort. In this work we investigate the requirements to create visualisations of execution traces of parallel programs modelled as dataflows. We propose the Smart Trace (ST) concept, to encode the structure of the data, and guide the construction of specialised visualizations. A visualization tool can then leverage the relationships in the data to automate a given analysis task. We show with examples the power and flexibility of visualisations we can create to address specific questions formulated about the analysis of the data, with emphasis in parallel dataflow traces.

Original languageEnglish (US)
Title of host publicationProceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014
PublisherIEEE Computer Society
Pages165-172
Number of pages8
ISBN (Print)9781479942602
DOIs
StatePublished - Oct 1 2014
Event2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014 - Rio de Janeiro, Brazil
Duration: Aug 27 2014Aug 30 2014

Other

Other2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014
CountryBrazil
CityRio de Janeiro
Period8/27/148/30/14

Fingerprint

Visualization

Keywords

  • Parallel computing
  • Trace visualization

ASJC Scopus subject areas

  • Computer Graphics and Computer-Aided Design
  • Signal Processing
  • Software

Cite this

Osmari, D. K., Vo, H. T., Silva, C. T., Comba, J. L. D., & Lins, L. (2014). Visualization and analysis of parallel dataflow execution with smart traces. In Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014 (pp. 165-172). [6915304] IEEE Computer Society. https://doi.org/10.1109/SIBGRAPI.2014.2

Visualization and analysis of parallel dataflow execution with smart traces. / Osmari, Daniel K.; Vo, Huy T.; Silva, Claudio T.; Comba, João L D; Lins, Lauro.

Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014. IEEE Computer Society, 2014. p. 165-172 6915304.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Osmari, DK, Vo, HT, Silva, CT, Comba, JLD & Lins, L 2014, Visualization and analysis of parallel dataflow execution with smart traces. in Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014., 6915304, IEEE Computer Society, pp. 165-172, 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014, Rio de Janeiro, Brazil, 8/27/14. https://doi.org/10.1109/SIBGRAPI.2014.2
Osmari DK, Vo HT, Silva CT, Comba JLD, Lins L. Visualization and analysis of parallel dataflow execution with smart traces. In Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014. IEEE Computer Society. 2014. p. 165-172. 6915304 https://doi.org/10.1109/SIBGRAPI.2014.2
Osmari, Daniel K. ; Vo, Huy T. ; Silva, Claudio T. ; Comba, João L D ; Lins, Lauro. / Visualization and analysis of parallel dataflow execution with smart traces. Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014. IEEE Computer Society, 2014. pp. 165-172
@inproceedings{8dee2449a3094f7f93581a1f38ae7481,
title = "Visualization and analysis of parallel dataflow execution with smart traces",
abstract = "Most performance analysis tools focus on presenting an overload of details, with little application-dependent structure, and predefined statistical summaries. This makes the complex relations present in a parallel program not directly recognisable to the user, making the task of identifying performance issues more costly in both time and effort. In this work we investigate the requirements to create visualisations of execution traces of parallel programs modelled as dataflows. We propose the Smart Trace (ST) concept, to encode the structure of the data, and guide the construction of specialised visualizations. A visualization tool can then leverage the relationships in the data to automate a given analysis task. We show with examples the power and flexibility of visualisations we can create to address specific questions formulated about the analysis of the data, with emphasis in parallel dataflow traces.",
keywords = "Parallel computing, Trace visualization",
author = "Osmari, {Daniel K.} and Vo, {Huy T.} and Silva, {Claudio T.} and Comba, {Jo{\~a}o L D} and Lauro Lins",
year = "2014",
month = "10",
day = "1",
doi = "10.1109/SIBGRAPI.2014.2",
language = "English (US)",
isbn = "9781479942602",
pages = "165--172",
booktitle = "Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014",
publisher = "IEEE Computer Society",

}

TY - GEN

T1 - Visualization and analysis of parallel dataflow execution with smart traces

AU - Osmari, Daniel K.

AU - Vo, Huy T.

AU - Silva, Claudio T.

AU - Comba, João L D

AU - Lins, Lauro

PY - 2014/10/1

Y1 - 2014/10/1

N2 - Most performance analysis tools focus on presenting an overload of details, with little application-dependent structure, and predefined statistical summaries. This makes the complex relations present in a parallel program not directly recognisable to the user, making the task of identifying performance issues more costly in both time and effort. In this work we investigate the requirements to create visualisations of execution traces of parallel programs modelled as dataflows. We propose the Smart Trace (ST) concept, to encode the structure of the data, and guide the construction of specialised visualizations. A visualization tool can then leverage the relationships in the data to automate a given analysis task. We show with examples the power and flexibility of visualisations we can create to address specific questions formulated about the analysis of the data, with emphasis in parallel dataflow traces.

AB - Most performance analysis tools focus on presenting an overload of details, with little application-dependent structure, and predefined statistical summaries. This makes the complex relations present in a parallel program not directly recognisable to the user, making the task of identifying performance issues more costly in both time and effort. In this work we investigate the requirements to create visualisations of execution traces of parallel programs modelled as dataflows. We propose the Smart Trace (ST) concept, to encode the structure of the data, and guide the construction of specialised visualizations. A visualization tool can then leverage the relationships in the data to automate a given analysis task. We show with examples the power and flexibility of visualisations we can create to address specific questions formulated about the analysis of the data, with emphasis in parallel dataflow traces.

KW - Parallel computing

KW - Trace visualization

UR - http://www.scopus.com/inward/record.url?scp=84908011631&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84908011631&partnerID=8YFLogxK

U2 - 10.1109/SIBGRAPI.2014.2

DO - 10.1109/SIBGRAPI.2014.2

M3 - Conference contribution

SN - 9781479942602

SP - 165

EP - 172

BT - Proceedings - 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images, SIBGRAPI 2014

PB - IEEE Computer Society

ER -