Formalizing bottlenecks in task-based OpenMP applications

Shajulin Benedict; Michael Gerndt; Diana Mihaela Gudu

doi:10.3233/978-1-61499-381-0-103

Formalizing bottlenecks in task-based OpenMP applications

Shajulin Benedict, Michael Gerndt, Diana Mihaela Gudu

Informatics 10 - Associate Professorship of Architecture of Parallel and Distributed Systems

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Task support was introduced into OpenMP to address irregular parallelism in shared memory architectures. Creating tasks that are extremely fine granular in applications, however, impedes performance. In this paper, a methodology for analyzing the performance of task-based OpenMP programs and its implementation in Periscope is presented. The paper unveils and concentrates on the newly formulated high-level performance properties that formalize typical performance bottlenecks of task-based programs. In addition, the paper reports on the experimental results which were accomplished for the codes of the Barcelona OpenMP Tasks Suite (BOTS) using Periscope in the SuperMUC supercomputing machine at Garching, Germany.

Original language	English
Title of host publication	Parallel Computing
Subtitle of host publication	Accelerating Computational Science and Engineering (CSE)
Publisher	IOS Press BV
Pages	103-112
Number of pages	10
ISBN (Print)	9781614993803
DOIs	https://doi.org/10.3233/978-1-61499-381-0-103
State	Published - 2014

Publication series

Name	Advances in Parallel Computing
Volume	25
ISSN (Print)	0927-5452

Keywords

OpenMP
Parallel Programming
Performance Analysis
Task Parallelism
Tools

Access to Document

10.3233/978-1-61499-381-0-103

Cite this

@inproceedings{efca683e8ad94f2892142426029d889e,

title = "Formalizing bottlenecks in task-based OpenMP applications",

abstract = "Task support was introduced into OpenMP to address irregular parallelism in shared memory architectures. Creating tasks that are extremely fine granular in applications, however, impedes performance. In this paper, a methodology for analyzing the performance of task-based OpenMP programs and its implementation in Periscope is presented. The paper unveils and concentrates on the newly formulated high-level performance properties that formalize typical performance bottlenecks of task-based programs. In addition, the paper reports on the experimental results which were accomplished for the codes of the Barcelona OpenMP Tasks Suite (BOTS) using Periscope in the SuperMUC supercomputing machine at Garching, Germany.",

keywords = "OpenMP, Parallel Programming, Performance Analysis, Task Parallelism, Tools",

author = "Shajulin Benedict and Michael Gerndt and Gudu, {Diana Mihaela}",

year = "2014",

doi = "10.3233/978-1-61499-381-0-103",

language = "English",

isbn = "9781614993803",

series = "Advances in Parallel Computing",

publisher = "IOS Press BV",

pages = "103--112",

booktitle = "Parallel Computing",

}

Formalizing bottlenecks in task-based OpenMP applications. / Benedict, Shajulin ; Gerndt, Michael; Gudu, Diana Mihaela.
Parallel Computing: Accelerating Computational Science and Engineering (CSE). IOS Press BV, 2014. p. 103-112 (Advances in Parallel Computing; Vol. 25).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Formalizing bottlenecks in task-based OpenMP applications

AU - Benedict, Shajulin

AU - Gerndt, Michael

AU - Gudu, Diana Mihaela

PY - 2014

Y1 - 2014

N2 - Task support was introduced into OpenMP to address irregular parallelism in shared memory architectures. Creating tasks that are extremely fine granular in applications, however, impedes performance. In this paper, a methodology for analyzing the performance of task-based OpenMP programs and its implementation in Periscope is presented. The paper unveils and concentrates on the newly formulated high-level performance properties that formalize typical performance bottlenecks of task-based programs. In addition, the paper reports on the experimental results which were accomplished for the codes of the Barcelona OpenMP Tasks Suite (BOTS) using Periscope in the SuperMUC supercomputing machine at Garching, Germany.

AB - Task support was introduced into OpenMP to address irregular parallelism in shared memory architectures. Creating tasks that are extremely fine granular in applications, however, impedes performance. In this paper, a methodology for analyzing the performance of task-based OpenMP programs and its implementation in Periscope is presented. The paper unveils and concentrates on the newly formulated high-level performance properties that formalize typical performance bottlenecks of task-based programs. In addition, the paper reports on the experimental results which were accomplished for the codes of the Barcelona OpenMP Tasks Suite (BOTS) using Periscope in the SuperMUC supercomputing machine at Garching, Germany.

KW - OpenMP

KW - Parallel Programming

KW - Performance Analysis

KW - Task Parallelism

KW - Tools

UR - http://www.scopus.com/inward/record.url?scp=84902275152&partnerID=8YFLogxK

U2 - 10.3233/978-1-61499-381-0-103

DO - 10.3233/978-1-61499-381-0-103

M3 - Conference contribution

AN - SCOPUS:84902275152

SN - 9781614993803

T3 - Advances in Parallel Computing

SP - 103

EP - 112

BT - Parallel Computing

PB - IOS Press BV

ER -

Formalizing bottlenecks in task-based OpenMP applications

Abstract

Publication series

Keywords

Access to Document

Other files and links

Fingerprint

Cite this