Cross-platform performance prediction of parallel applications using partial execution

Leo T. Yang, Xiaosong Ma, Frank Mueller

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

87 Citations (Scopus)

Abstract

Performance prediction across platforms is increasingly important as developers can choose from a wide range of execution platforms. The main challenge remains to perform accurate predictions at a low-cost across different architectures. In this paper, we derive an affordable method approaching cross-platform performance translation based on relative performance between two platforms. We argue that relative performance can be observed without running a parallel application in full. We show that it suffices to observe very short partial executions of an application since most parallel codes are iterative and behave predictably manner after a minimal startup period. This novel prediction approach is observation-based. It does not require program modeling, code analysis, or architectural simulation. Our performance results using real platforms and production codes demonstrate that prediction derived from partial executions can yield high accuracy at a low cost. We also assess the limitations of our model and identify future research directions on observation-based performance prediction.

Original languageEnglish
Title of host publicationProceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05
Volume2005
DOIs
Publication statusPublished - 1 Dec 2005
Externally publishedYes
EventACM/IEEE 2005 Supercomputing Conference, SC'05 - Seatle, WA, United States
Duration: 12 Nov 200518 Nov 2005

Other

OtherACM/IEEE 2005 Supercomputing Conference, SC'05
CountryUnited States
CitySeatle, WA
Period12/11/0518/11/05

Fingerprint

Costs

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Yang, L. T., Ma, X., & Mueller, F. (2005). Cross-platform performance prediction of parallel applications using partial execution. In Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05 (Vol. 2005). [1559992] https://doi.org/10.1109/SC.2005.20

Cross-platform performance prediction of parallel applications using partial execution. / Yang, Leo T.; Ma, Xiaosong; Mueller, Frank.

Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05. Vol. 2005 2005. 1559992.

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

Yang, LT, Ma, X & Mueller, F 2005, Cross-platform performance prediction of parallel applications using partial execution. in Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05. vol. 2005, 1559992, ACM/IEEE 2005 Supercomputing Conference, SC'05, Seatle, WA, United States, 12/11/05. https://doi.org/10.1109/SC.2005.20
Yang LT, Ma X, Mueller F. Cross-platform performance prediction of parallel applications using partial execution. In Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05. Vol. 2005. 2005. 1559992 https://doi.org/10.1109/SC.2005.20
Yang, Leo T. ; Ma, Xiaosong ; Mueller, Frank. / Cross-platform performance prediction of parallel applications using partial execution. Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05. Vol. 2005 2005.
@inproceedings{eefb1fe4b2db4fa8ab2448b1fd5a7bd8,
title = "Cross-platform performance prediction of parallel applications using partial execution",
abstract = "Performance prediction across platforms is increasingly important as developers can choose from a wide range of execution platforms. The main challenge remains to perform accurate predictions at a low-cost across different architectures. In this paper, we derive an affordable method approaching cross-platform performance translation based on relative performance between two platforms. We argue that relative performance can be observed without running a parallel application in full. We show that it suffices to observe very short partial executions of an application since most parallel codes are iterative and behave predictably manner after a minimal startup period. This novel prediction approach is observation-based. It does not require program modeling, code analysis, or architectural simulation. Our performance results using real platforms and production codes demonstrate that prediction derived from partial executions can yield high accuracy at a low cost. We also assess the limitations of our model and identify future research directions on observation-based performance prediction.",
author = "Yang, {Leo T.} and Xiaosong Ma and Frank Mueller",
year = "2005",
month = "12",
day = "1",
doi = "10.1109/SC.2005.20",
language = "English",
isbn = "1595930612",
volume = "2005",
booktitle = "Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05",

}

TY - GEN

T1 - Cross-platform performance prediction of parallel applications using partial execution

AU - Yang, Leo T.

AU - Ma, Xiaosong

AU - Mueller, Frank

PY - 2005/12/1

Y1 - 2005/12/1

N2 - Performance prediction across platforms is increasingly important as developers can choose from a wide range of execution platforms. The main challenge remains to perform accurate predictions at a low-cost across different architectures. In this paper, we derive an affordable method approaching cross-platform performance translation based on relative performance between two platforms. We argue that relative performance can be observed without running a parallel application in full. We show that it suffices to observe very short partial executions of an application since most parallel codes are iterative and behave predictably manner after a minimal startup period. This novel prediction approach is observation-based. It does not require program modeling, code analysis, or architectural simulation. Our performance results using real platforms and production codes demonstrate that prediction derived from partial executions can yield high accuracy at a low cost. We also assess the limitations of our model and identify future research directions on observation-based performance prediction.

AB - Performance prediction across platforms is increasingly important as developers can choose from a wide range of execution platforms. The main challenge remains to perform accurate predictions at a low-cost across different architectures. In this paper, we derive an affordable method approaching cross-platform performance translation based on relative performance between two platforms. We argue that relative performance can be observed without running a parallel application in full. We show that it suffices to observe very short partial executions of an application since most parallel codes are iterative and behave predictably manner after a minimal startup period. This novel prediction approach is observation-based. It does not require program modeling, code analysis, or architectural simulation. Our performance results using real platforms and production codes demonstrate that prediction derived from partial executions can yield high accuracy at a low cost. We also assess the limitations of our model and identify future research directions on observation-based performance prediction.

UR - http://www.scopus.com/inward/record.url?scp=33845442055&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33845442055&partnerID=8YFLogxK

U2 - 10.1109/SC.2005.20

DO - 10.1109/SC.2005.20

M3 - Conference contribution

AN - SCOPUS:33845442055

SN - 1595930612

SN - 9781595930613

VL - 2005

BT - Proceedings of the ACM/IEEE 2005 Supercomputing Conference, SC'05

ER -