A fault-tolerance protocol for parallel applications with communication imbalance

Esteban Meneses; Laxmikant V. Kale

doi:10.1109/SBAC-PAD.2015.25

A fault-tolerance protocol for parallel applications with communication imbalance

Esteban Meneses
, Laxmikant V. Kale

Dept. of Computer Science, University of Illinois at Urbana-Champaign

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

Abstract

The predicted failure rates of future supercomputers loom the groundbreaking research large machines are expected to foster. Therefore, resilient extreme-scale applications are an absolute necessity to effectively use the new generation of supercomputers. Rollback-recovery techniques have been traditionally used in HPC to provide resilience. Among those techniques, message logging provides the appealing features of saving energy, accelerating recovery, and having low performance penalty. Its increased memory consumption is, however, an important downside. This paper introduces memory-constrained message logging (MCML), a general framework for decreasing the memory footprint of message-logging protocols. In particular, we demonstrate the effectiveness of MCML in maintaining message logging feasible for applications with substantial communication imbalance. This type of applications appear in many scientific fields. We present experimental results with several parallel codes running on up to 4,096 cores. Using those results and an analytical model, we predict MCML can reduce execution time up to 25% and energy consumption up to 15%, at extreme scale.

Original language	English
Title of host publication	Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015
Publisher	IEEE Computer Society
Pages	162-169
Number of pages	8
ISBN (Electronic)	9781467380119
DOIs	https://doi.org/10.1109/SBAC-PAD.2015.25
State	Published - 12 Jan 2016
Event	27th IEEE International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015 - Florianopolis, Brazil Duration: 18 Oct 2015 → 21 Oct 2015

Publication series

Name	Proceedings - Symposium on Computer Architecture and High Performance Computing
Volume	2016-January
ISSN (Print)	1550-6533

Conference

Conference	27th IEEE International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015
Country/Territory	Brazil
City	Florianopolis
Period	18/10/15 → 21/10/15

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

Communication imbalance
Fault tolerance
Message logging

Access to Document

10.1109/SBAC-PAD.2015.25

Cite this

Meneses, E., & Kale, L. V. (2016). A fault-tolerance protocol for parallel applications with communication imbalance. In Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015 (pp. 162-169). Article 7379847 (Proceedings - Symposium on Computer Architecture and High Performance Computing; Vol. 2016-January). IEEE Computer Society. https://doi.org/10.1109/SBAC-PAD.2015.25

@inproceedings{e36634e5d9ed44149a458398a63d27a3,

title = "A fault-tolerance protocol for parallel applications with communication imbalance",

abstract = "The predicted failure rates of future supercomputers loom the groundbreaking research large machines are expected to foster. Therefore, resilient extreme-scale applications are an absolute necessity to effectively use the new generation of supercomputers. Rollback-recovery techniques have been traditionally used in HPC to provide resilience. Among those techniques, message logging provides the appealing features of saving energy, accelerating recovery, and having low performance penalty. Its increased memory consumption is, however, an important downside. This paper introduces memory-constrained message logging (MCML), a general framework for decreasing the memory footprint of message-logging protocols. In particular, we demonstrate the effectiveness of MCML in maintaining message logging feasible for applications with substantial communication imbalance. This type of applications appear in many scientific fields. We present experimental results with several parallel codes running on up to 4,096 cores. Using those results and an analytical model, we predict MCML can reduce execution time up to 25\% and energy consumption up to 15\%, at extreme scale.",

keywords = "Communication imbalance, Fault tolerance, Message logging",

author = "Esteban Meneses and Kale, \{Laxmikant V.\}",

note = "Publisher Copyright: {\textcopyright} 2015 IEEE.; 27th IEEE International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015 ; Conference date: 18-10-2015 Through 21-10-2015",

year = "2016",

month = jan,

day = "12",

doi = "10.1109/SBAC-PAD.2015.25",

language = "Ingl{\'e}s",

series = "Proceedings - Symposium on Computer Architecture and High Performance Computing",

publisher = "IEEE Computer Society",

pages = "162--169",

booktitle = "Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015",

}

Meneses, E & Kale, LV 2016, A fault-tolerance protocol for parallel applications with communication imbalance. in Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015., 7379847, Proceedings - Symposium on Computer Architecture and High Performance Computing, vol. 2016-January, IEEE Computer Society, pp. 162-169, 27th IEEE International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015, Florianopolis, Brazil, 18/10/15. https://doi.org/10.1109/SBAC-PAD.2015.25

A fault-tolerance protocol for parallel applications with communication imbalance. / Meneses, Esteban; Kale, Laxmikant V.
Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015. IEEE Computer Society, 2016. p. 162-169 7379847 (Proceedings - Symposium on Computer Architecture and High Performance Computing; Vol. 2016-January).

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

TY - GEN

T1 - A fault-tolerance protocol for parallel applications with communication imbalance

AU - Meneses, Esteban

AU - Kale, Laxmikant V.

PY - 2016/1/12

Y1 - 2016/1/12

N2 - The predicted failure rates of future supercomputers loom the groundbreaking research large machines are expected to foster. Therefore, resilient extreme-scale applications are an absolute necessity to effectively use the new generation of supercomputers. Rollback-recovery techniques have been traditionally used in HPC to provide resilience. Among those techniques, message logging provides the appealing features of saving energy, accelerating recovery, and having low performance penalty. Its increased memory consumption is, however, an important downside. This paper introduces memory-constrained message logging (MCML), a general framework for decreasing the memory footprint of message-logging protocols. In particular, we demonstrate the effectiveness of MCML in maintaining message logging feasible for applications with substantial communication imbalance. This type of applications appear in many scientific fields. We present experimental results with several parallel codes running on up to 4,096 cores. Using those results and an analytical model, we predict MCML can reduce execution time up to 25% and energy consumption up to 15%, at extreme scale.

AB - The predicted failure rates of future supercomputers loom the groundbreaking research large machines are expected to foster. Therefore, resilient extreme-scale applications are an absolute necessity to effectively use the new generation of supercomputers. Rollback-recovery techniques have been traditionally used in HPC to provide resilience. Among those techniques, message logging provides the appealing features of saving energy, accelerating recovery, and having low performance penalty. Its increased memory consumption is, however, an important downside. This paper introduces memory-constrained message logging (MCML), a general framework for decreasing the memory footprint of message-logging protocols. In particular, we demonstrate the effectiveness of MCML in maintaining message logging feasible for applications with substantial communication imbalance. This type of applications appear in many scientific fields. We present experimental results with several parallel codes running on up to 4,096 cores. Using those results and an analytical model, we predict MCML can reduce execution time up to 25% and energy consumption up to 15%, at extreme scale.

KW - Communication imbalance

KW - Fault tolerance

KW - Message logging

UR - https://www.scopus.com/pages/publications/84964903768

U2 - 10.1109/SBAC-PAD.2015.25

DO - 10.1109/SBAC-PAD.2015.25

M3 - Contribución a la conferencia

AN - SCOPUS:84964903768

T3 - Proceedings - Symposium on Computer Architecture and High Performance Computing

SP - 162

EP - 169

BT - Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015

PB - IEEE Computer Society

T2 - 27th IEEE International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015

Y2 - 18 October 2015 through 21 October 2015

ER -

Meneses E, Kale LV. A fault-tolerance protocol for parallel applications with communication imbalance. In Proceedings - IEEE 27th International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2015. IEEE Computer Society. 2016. p. 162-169. 7379847. (Proceedings - Symposium on Computer Architecture and High Performance Computing). doi: 10.1109/SBAC-PAD.2015.25

A fault-tolerance protocol for parallel applications with communication imbalance

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this