A robust dynamic optimization for MPI alltoall operation

Hyacinthe Nzigou Mamadou, Takeshi Nanri, Kazuaki Murakami

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

2 Citations (Scopus)

Abstract

The performance of the Message Passing Interface collective communications is a critical issue to high performance computing widely discussed. In this paper we propose a mechanism that dynamically selects the most efficient MPI Alltoall algorithm for a given system/workload situation. This implementation method starts by grouping the fast algorithms based on respective performance prediction models that were obtained by using the point-to-point model P-LogP. The experiments performed on different parallel machines equipped with Infiniband and Gigabit Ethernet interconnects produced encouraging results, with negligible overhead to find the most appropriate algorithm to carry on the operation. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.

Original languageEnglish
Title of host publicationIPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium
DOIs
Publication statusPublished - Nov 25 2009
Event23rd IEEE International Parallel and Distributed Processing Symposium, IPDPS 2009 - Rome, Italy
Duration: May 23 2009May 29 2009

Publication series

NameIPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium

Other

Other23rd IEEE International Parallel and Distributed Processing Symposium, IPDPS 2009
CountryItaly
CityRome
Period5/23/095/29/09

Fingerprint

Message passing
Ethernet
Communication
Experiments

All Science Journal Classification (ASJC) codes

  • Computational Theory and Mathematics
  • Hardware and Architecture
  • Software

Cite this

Mamadou, H. N., Nanri, T., & Murakami, K. (2009). A robust dynamic optimization for MPI alltoall operation. In IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium [5160917] (IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium). https://doi.org/10.1109/IPDPS.2009.5160917

A robust dynamic optimization for MPI alltoall operation. / Mamadou, Hyacinthe Nzigou; Nanri, Takeshi; Murakami, Kazuaki.

IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium. 2009. 5160917 (IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium).

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

Mamadou, HN, Nanri, T & Murakami, K 2009, A robust dynamic optimization for MPI alltoall operation. in IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium., 5160917, IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium, 23rd IEEE International Parallel and Distributed Processing Symposium, IPDPS 2009, Rome, Italy, 5/23/09. https://doi.org/10.1109/IPDPS.2009.5160917
Mamadou HN, Nanri T, Murakami K. A robust dynamic optimization for MPI alltoall operation. In IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium. 2009. 5160917. (IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium). https://doi.org/10.1109/IPDPS.2009.5160917
Mamadou, Hyacinthe Nzigou ; Nanri, Takeshi ; Murakami, Kazuaki. / A robust dynamic optimization for MPI alltoall operation. IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium. 2009. (IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium).
@inproceedings{c7655042164b4d1db95e81a9dc8410cb,
title = "A robust dynamic optimization for MPI alltoall operation",
abstract = "The performance of the Message Passing Interface collective communications is a critical issue to high performance computing widely discussed. In this paper we propose a mechanism that dynamically selects the most efficient MPI Alltoall algorithm for a given system/workload situation. This implementation method starts by grouping the fast algorithms based on respective performance prediction models that were obtained by using the point-to-point model P-LogP. The experiments performed on different parallel machines equipped with Infiniband and Gigabit Ethernet interconnects produced encouraging results, with negligible overhead to find the most appropriate algorithm to carry on the operation. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.",
author = "Mamadou, {Hyacinthe Nzigou} and Takeshi Nanri and Kazuaki Murakami",
year = "2009",
month = "11",
day = "25",
doi = "10.1109/IPDPS.2009.5160917",
language = "English",
isbn = "9781424437504",
series = "IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium",
booktitle = "IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium",

}

TY - GEN

T1 - A robust dynamic optimization for MPI alltoall operation

AU - Mamadou, Hyacinthe Nzigou

AU - Nanri, Takeshi

AU - Murakami, Kazuaki

PY - 2009/11/25

Y1 - 2009/11/25

N2 - The performance of the Message Passing Interface collective communications is a critical issue to high performance computing widely discussed. In this paper we propose a mechanism that dynamically selects the most efficient MPI Alltoall algorithm for a given system/workload situation. This implementation method starts by grouping the fast algorithms based on respective performance prediction models that were obtained by using the point-to-point model P-LogP. The experiments performed on different parallel machines equipped with Infiniband and Gigabit Ethernet interconnects produced encouraging results, with negligible overhead to find the most appropriate algorithm to carry on the operation. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.

AB - The performance of the Message Passing Interface collective communications is a critical issue to high performance computing widely discussed. In this paper we propose a mechanism that dynamically selects the most efficient MPI Alltoall algorithm for a given system/workload situation. This implementation method starts by grouping the fast algorithms based on respective performance prediction models that were obtained by using the point-to-point model P-LogP. The experiments performed on different parallel machines equipped with Infiniband and Gigabit Ethernet interconnects produced encouraging results, with negligible overhead to find the most appropriate algorithm to carry on the operation. In most cases, the dynamic Alltoall largely outperforms the traditional MPI implementations on different platforms.

UR - http://www.scopus.com/inward/record.url?scp=70449793546&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70449793546&partnerID=8YFLogxK

U2 - 10.1109/IPDPS.2009.5160917

DO - 10.1109/IPDPS.2009.5160917

M3 - Conference contribution

AN - SCOPUS:70449793546

SN - 9781424437504

T3 - IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium

BT - IPDPS 2009 - Proceedings of the 2009 IEEE International Parallel and Distributed Processing Symposium

ER -