Automatic analysis of inefficiency patterns in parallel applications

Felix Wolf; Bernd Mohr; Jack Dongarra; Shirley Moore

doi:10.1002/cpe.1128

Automatic analysis of inefficiency patterns in parallel applications

Felix Wolf, Bernd Mohr, Jack Dongarra, Shirley Moore

Research output: Contribution to journal › Article › peer-review

11 Scopus citations

Abstract

Event tracing is a powerful method for analyzing the performance behavior of parallel applications. Because event traces record the temporal and spatial relationships between individual runtime events, they allow application developers to analyze dependences of performance phenomena across concurrent control flows. However, in view of the large amounts of data generated on contemporary parallel machines, the depth and coverage of a purely manual analysis is often limited. Our approach automatically searches event traces for patterns of inefficient behavior, classifies detected instances by category, and quantifies the associated performance penalty. This enables developers to study the performance of their applications at a high level of abstraction, while requiring significantly less time and expertise than a manual analysis.

Original language	English
Pages (from-to)	1481-1496
Number of pages	16
Journal	Concurrency and Computation: Practice and Experience
Volume	19
Issue number	11
DOIs	https://doi.org/10.1002/cpe.1128
State	Published - Aug 10 2007
Externally published	Yes

Keywords

Event tracing
Pattern search
Performance tools

Access to Document

10.1002/cpe.1128

Cite this

@article{c6c08bb9674343c2a68a02a2c244f0e6,

title = "Automatic analysis of inefficiency patterns in parallel applications",

abstract = "Event tracing is a powerful method for analyzing the performance behavior of parallel applications. Because event traces record the temporal and spatial relationships between individual runtime events, they allow application developers to analyze dependences of performance phenomena across concurrent control flows. However, in view of the large amounts of data generated on contemporary parallel machines, the depth and coverage of a purely manual analysis is often limited. Our approach automatically searches event traces for patterns of inefficient behavior, classifies detected instances by category, and quantifies the associated performance penalty. This enables developers to study the performance of their applications at a high level of abstraction, while requiring significantly less time and expertise than a manual analysis.",

keywords = "Event tracing, Pattern search, Performance tools",

author = "Felix Wolf and Bernd Mohr and Jack Dongarra and Shirley Moore",

year = "2007",

month = aug,

day = "10",

doi = "10.1002/cpe.1128",

language = "English",

volume = "19",

pages = "1481--1496",

journal = "Concurrency and Computation: Practice and Experience",

issn = "1532-0626",

publisher = "wiley",

number = "11",

}

TY - JOUR

T1 - Automatic analysis of inefficiency patterns in parallel applications

AU - Wolf, Felix

AU - Mohr, Bernd

AU - Dongarra, Jack

AU - Moore, Shirley

PY - 2007/8/10

Y1 - 2007/8/10

N2 - Event tracing is a powerful method for analyzing the performance behavior of parallel applications. Because event traces record the temporal and spatial relationships between individual runtime events, they allow application developers to analyze dependences of performance phenomena across concurrent control flows. However, in view of the large amounts of data generated on contemporary parallel machines, the depth and coverage of a purely manual analysis is often limited. Our approach automatically searches event traces for patterns of inefficient behavior, classifies detected instances by category, and quantifies the associated performance penalty. This enables developers to study the performance of their applications at a high level of abstraction, while requiring significantly less time and expertise than a manual analysis.

AB - Event tracing is a powerful method for analyzing the performance behavior of parallel applications. Because event traces record the temporal and spatial relationships between individual runtime events, they allow application developers to analyze dependences of performance phenomena across concurrent control flows. However, in view of the large amounts of data generated on contemporary parallel machines, the depth and coverage of a purely manual analysis is often limited. Our approach automatically searches event traces for patterns of inefficient behavior, classifies detected instances by category, and quantifies the associated performance penalty. This enables developers to study the performance of their applications at a high level of abstraction, while requiring significantly less time and expertise than a manual analysis.

KW - Event tracing

KW - Pattern search

KW - Performance tools

UR - http://www.scopus.com/inward/record.url?scp=34547616149&partnerID=8YFLogxK

U2 - 10.1002/cpe.1128

DO - 10.1002/cpe.1128

M3 - Article

AN - SCOPUS:34547616149

SN - 1532-0626

VL - 19

SP - 1481

EP - 1496

JO - Concurrency and Computation: Practice and Experience

JF - Concurrency and Computation: Practice and Experience

IS - 11

ER -

Automatic analysis of inefficiency patterns in parallel applications

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this