Performance analysis of MPI collective operations

Jelena Pješivac-Grbović; Thara Angskun; George Bosilca; Graham E. Fagg; Edgar Gabriel; Jack J. Dongarra

doi:10.1109/IPDPS.2005.335

Performance analysis of MPI collective operations

Jelena Pješivac-Grbović, Thara Angskun, George Bosilca, Graham E. Fagg, Edgar Gabriel, Jack J. Dongarra

Research output: Chapter in Book/Conference proceeding › Conference contribution

Abstract

Previous studies of application usage show that the performance of collective communications are critical for high-performance computing and are often overlooked when compared to the point-to-point performance. In this paper, we analyze and attempt to improve infra-cluster collective communication in the context of the widely deployed MPI programming paradigm by extending accepted models of point-to-point communication, such as Hockney, LogP/LogGP, and PLogP. The predictions from the models were compared to the experimentally gathered data and our findings were used to optimize the implementation of collective operations in the FT-MPI library.

Original language	English
Title of host publication	Proceedings - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005\|Proc. 19th IEEE Int. Parallel Distr. Proces. Symp.
Publisher	IEEE Computer Society
Volume	2005
ISBN (Print)	0769523129, 9780769523125
DOIs	https://doi.org/10.1109/IPDPS.2005.335
Publication status	Published - 2005
Event	19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005 - Denver, CO Duration: 1 Jul 2005 → …

Conference

Conference	19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005
City	Denver, CO
Period	1/07/05 → …

Access to Document

10.1109/IPDPS.2005.335

Cite this

Pješivac-Grbović, J., Angskun, T., Bosilca, G., Fagg, G. E., Gabriel, E., & Dongarra, J. J. (2005). Performance analysis of MPI collective operations. In Proceedings - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005|Proc. 19th IEEE Int. Parallel Distr. Proces. Symp. (Vol. 2005). IEEE Computer Society . https://doi.org/10.1109/IPDPS.2005.335

@inproceedings{fd217f4df1f44cce8404255948da05f1,

title = "Performance analysis of MPI collective operations",

abstract = "Previous studies of application usage show that the performance of collective communications are critical for high-performance computing and are often overlooked when compared to the point-to-point performance. In this paper, we analyze and attempt to improve infra-cluster collective communication in the context of the widely deployed MPI programming paradigm by extending accepted models of point-to-point communication, such as Hockney, LogP/LogGP, and PLogP. The predictions from the models were compared to the experimentally gathered data and our findings were used to optimize the implementation of collective operations in the FT-MPI library.",

author = "Jelena Pje{\v s}ivac-Grbovi{\'c} and Thara Angskun and George Bosilca and Fagg, {Graham E.} and Edgar Gabriel and Dongarra, {Jack J.}",

year = "2005",

doi = "10.1109/IPDPS.2005.335",

language = "English",

isbn = "0769523129",

volume = "2005",

booktitle = "Proceedings - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005|Proc. 19th IEEE Int. Parallel Distr. Proces. Symp.",

publisher = "IEEE Computer Society ",

address = "United States",

note = "19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005 ; Conference date: 01-07-2005",

}

Pješivac-Grbović, J, Angskun, T, Bosilca, G, Fagg, GE, Gabriel, E & Dongarra, JJ 2005, Performance analysis of MPI collective operations. in Proceedings - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005|Proc. 19th IEEE Int. Parallel Distr. Proces. Symp.. vol. 2005, IEEE Computer Society , 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005, Denver, CO, 1/07/05. https://doi.org/10.1109/IPDPS.2005.335

Performance analysis of MPI collective operations. / Pješivac-Grbović, Jelena; Angskun, Thara; Bosilca, George et al.
Proceedings - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005|Proc. 19th IEEE Int. Parallel Distr. Proces. Symp.. Vol. 2005 IEEE Computer Society , 2005.

Research output: Chapter in Book/Conference proceeding › Conference contribution

TY - GEN

T1 - Performance analysis of MPI collective operations

AU - Pješivac-Grbović, Jelena

AU - Angskun, Thara

AU - Bosilca, George

AU - Fagg, Graham E.

AU - Gabriel, Edgar

AU - Dongarra, Jack J.

PY - 2005

Y1 - 2005

N2 - Previous studies of application usage show that the performance of collective communications are critical for high-performance computing and are often overlooked when compared to the point-to-point performance. In this paper, we analyze and attempt to improve infra-cluster collective communication in the context of the widely deployed MPI programming paradigm by extending accepted models of point-to-point communication, such as Hockney, LogP/LogGP, and PLogP. The predictions from the models were compared to the experimentally gathered data and our findings were used to optimize the implementation of collective operations in the FT-MPI library.

AB - Previous studies of application usage show that the performance of collective communications are critical for high-performance computing and are often overlooked when compared to the point-to-point performance. In this paper, we analyze and attempt to improve infra-cluster collective communication in the context of the widely deployed MPI programming paradigm by extending accepted models of point-to-point communication, such as Hockney, LogP/LogGP, and PLogP. The predictions from the models were compared to the experimentally gathered data and our findings were used to optimize the implementation of collective operations in the FT-MPI library.

U2 - 10.1109/IPDPS.2005.335

DO - 10.1109/IPDPS.2005.335

M3 - Conference contribution

SN - 0769523129

SN - 9780769523125

VL - 2005

BT - Proceedings - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005|Proc. 19th IEEE Int. Parallel Distr. Proces. Symp.

PB - IEEE Computer Society

T2 - 19th IEEE International Parallel and Distributed Processing Symposium, IPDPS 2005

Y2 - 1 July 2005

ER -

Performance analysis of MPI collective operations

Abstract

Conference

Access to Document

Fingerprint

Cite this