A dynamic solution for efficient MPI collective communications

Hyacinthe Nzigou Mamadou, Feng Long Gu, Vivien Oddou, Takeshi Nanri, Kazuaki Murakami

研究成果: 著書/レポートタイプへの貢献会議での発言

抄録

The performance of the Message Passing Interface collective communication algorithms is a critical issue widely discussed in both academia and industry. In order to achieve and maintain high performance in the MPI implementations even in regard to random system behavior, the collective operations must be adapted to both the cluster platform and the workload of the user program. In this paper we propose the DYN Alltoall, a dynamic version of the traditional MPI Alltoall implementation, which is based on performance predictions derived from P-LogP model. The experiments which were performed on clusters equipped with different interconnect networks, Infiniband and Gigabit Ethernet, produced encouraging results with negligible overhead to find the most appropriate algorithm. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.

元の言語英語
ホスト出版物のタイトルProceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009
ページ3-7
ページ数5
DOI
出版物ステータス出版済み - 12 2 2009
イベント2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009 - Sanya, Hainan, 中国
継続期間: 4 24 20094 26 2009

出版物シリーズ

名前Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009
1

その他

その他2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009
中国
Sanya, Hainan
期間4/24/094/26/09

Fingerprint

Collective Communication
Communication
Message passing
Ethernet
InfiniBand
Random Systems
Message Passing Interface
Performance Prediction
Interconnect
Workload
High Performance
Industry
Experiments
Experiment
Model

All Science Journal Classification (ASJC) codes

  • Computational Theory and Mathematics
  • Computer Science Applications
  • Theoretical Computer Science

これを引用

Mamadou, H. N., Gu, F. L., Oddou, V., Nanri, T., & Murakami, K. (2009). A dynamic solution for efficient MPI collective communications. : Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009 (pp. 3-7). [5193631] (Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009; 巻数 1). https://doi.org/10.1109/CSO.2009.448

A dynamic solution for efficient MPI collective communications. / Mamadou, Hyacinthe Nzigou; Gu, Feng Long; Oddou, Vivien; Nanri, Takeshi; Murakami, Kazuaki.

Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009. 2009. p. 3-7 5193631 (Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009; 巻 1).

研究成果: 著書/レポートタイプへの貢献会議での発言

Mamadou, HN, Gu, FL, Oddou, V, Nanri, T & Murakami, K 2009, A dynamic solution for efficient MPI collective communications. : Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009., 5193631, Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009, 巻. 1, pp. 3-7, 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009, Sanya, Hainan, 中国, 4/24/09. https://doi.org/10.1109/CSO.2009.448
Mamadou HN, Gu FL, Oddou V, Nanri T, Murakami K. A dynamic solution for efficient MPI collective communications. : Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009. 2009. p. 3-7. 5193631. (Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009). https://doi.org/10.1109/CSO.2009.448
Mamadou, Hyacinthe Nzigou ; Gu, Feng Long ; Oddou, Vivien ; Nanri, Takeshi ; Murakami, Kazuaki. / A dynamic solution for efficient MPI collective communications. Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009. 2009. pp. 3-7 (Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009).
@inproceedings{0bede9b6127d4a5299a67c3c766d0a26,
title = "A dynamic solution for efficient MPI collective communications",
abstract = "The performance of the Message Passing Interface collective communication algorithms is a critical issue widely discussed in both academia and industry. In order to achieve and maintain high performance in the MPI implementations even in regard to random system behavior, the collective operations must be adapted to both the cluster platform and the workload of the user program. In this paper we propose the DYN Alltoall, a dynamic version of the traditional MPI Alltoall implementation, which is based on performance predictions derived from P-LogP model. The experiments which were performed on clusters equipped with different interconnect networks, Infiniband and Gigabit Ethernet, produced encouraging results with negligible overhead to find the most appropriate algorithm. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.",
author = "Mamadou, {Hyacinthe Nzigou} and Gu, {Feng Long} and Vivien Oddou and Takeshi Nanri and Kazuaki Murakami",
year = "2009",
month = "12",
day = "2",
doi = "10.1109/CSO.2009.448",
language = "English",
isbn = "9780769536057",
series = "Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009",
pages = "3--7",
booktitle = "Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009",

}

TY - GEN

T1 - A dynamic solution for efficient MPI collective communications

AU - Mamadou, Hyacinthe Nzigou

AU - Gu, Feng Long

AU - Oddou, Vivien

AU - Nanri, Takeshi

AU - Murakami, Kazuaki

PY - 2009/12/2

Y1 - 2009/12/2

N2 - The performance of the Message Passing Interface collective communication algorithms is a critical issue widely discussed in both academia and industry. In order to achieve and maintain high performance in the MPI implementations even in regard to random system behavior, the collective operations must be adapted to both the cluster platform and the workload of the user program. In this paper we propose the DYN Alltoall, a dynamic version of the traditional MPI Alltoall implementation, which is based on performance predictions derived from P-LogP model. The experiments which were performed on clusters equipped with different interconnect networks, Infiniband and Gigabit Ethernet, produced encouraging results with negligible overhead to find the most appropriate algorithm. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.

AB - The performance of the Message Passing Interface collective communication algorithms is a critical issue widely discussed in both academia and industry. In order to achieve and maintain high performance in the MPI implementations even in regard to random system behavior, the collective operations must be adapted to both the cluster platform and the workload of the user program. In this paper we propose the DYN Alltoall, a dynamic version of the traditional MPI Alltoall implementation, which is based on performance predictions derived from P-LogP model. The experiments which were performed on clusters equipped with different interconnect networks, Infiniband and Gigabit Ethernet, produced encouraging results with negligible overhead to find the most appropriate algorithm. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.

UR - http://www.scopus.com/inward/record.url?scp=70649100401&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70649100401&partnerID=8YFLogxK

U2 - 10.1109/CSO.2009.448

DO - 10.1109/CSO.2009.448

M3 - Conference contribution

AN - SCOPUS:70649100401

SN - 9780769536057

T3 - Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009

SP - 3

EP - 7

BT - Proceedings of the 2009 International Joint Conference on Computational Sciences and Optimization, CSO 2009

ER -