F. Desprez,
P. Ramet,
and J. Roman.
Optimal Grain Size Computation for Pipelined Algorithms.
In Proceedings of Euro-Par'96,
number 1123 of LNCS,
Lyon, France,
pages 165-172,
1996.
Springer Verlag.
Keyword(s): Overlap.
Abstract: |
In this paper, we present a method for overlapping communications on parallel computers for pipelined algorithms. We first introduce a general theoretical model which leads to a generic computation scheme for the optimal packet size. Then, we use the OPIUM library, which provides an easy-to-use and efficient way to compute, in the general case, this optimal packet size, on the column $LU$ factorization; the implementation and performance measures are made on an Intel Paragon. |
@InProceedings{C:LaBRI::PR1,
author = "Desprez, F. and Ramet, P. and Roman, J.",
title = "Optimal Grain Size Computation for Pipelined Algorithms",
OPTcrossref = "",
OPTkey = "",
OPTeditor = "",
OPTvolume = "",
number = "1123",
series = "LNCS",
pages = "165-172",
booktitle = "Proceedings of Euro-Par'96",
year = "1996",
OPTorganization = "",
publisher = "Springer Verlag",
address = "Lyon, France",
OPTmonth = "",
OPTnote = "",
OPTannote = "",
URL = {http://www.labri.fr/~ramet/restricted/europar96.ps},
KEYWORDS = "Overlap",
ABSTRACT = {In this paper, we present a method for overlapping communications on parallel computers for pipelined algorithms. We first introduce a general theoretical model which leads to a generic computation scheme for the optimal packet size. Then, we use the OPIUM library, which provides an easy-to-use and efficient way to compute, in the general case, this optimal packet size, on the column $LU$ factorization; the implementation and performance measures are made on an Intel Paragon.}
}
P. Ramet.
Calcul de la taille optimale des paquets pour les algorithmes macro-pipelines.
In ACTES RenPar'8,
Bordeaux, France,
pages 21-24,
1996.
Keyword(s): Overlap.
Abstract: |
Le cadre g\'en\'eral de ce travail concerne les machines parall\`eles M.I.M.D. \`a m\'emoire distribu\'ee. Dans ce contexte, il est admis que la bonne d\'emarche de gestion du parall\'elisme passe par l'utilisation syst\'ematique de bibliot\`eques performantes de calcul et de communication avec la possibilit\'e de faire du recouvrement pour masquer, dans le temps, les communications par des calculs. Le th\`eme central de ce travail concerne le recouvrement calcul/communication, avec en particulier le calcul, \'eventuellement adaptatif, de la taille optimale des paquets \`a communiquer. |
@InProceedings{c:LaBRI::PR2,
author = "Ramet, P.",
title = "Calcul de la taille optimale des paquets pour les algorithmes macro-pipelines",
OPTcrossref = "",
OPTkey = "",
OPTeditor = "",
OPTvolume = "",
OPTnumber = "",
OPTseries = "",
pages = "21-24",
booktitle = "ACTES RenPar'8",
year = "1996",
OPTorganization = "",
OPTpublisher = "",
address = "Bordeaux, France",
OPTmonth = "",
OPTnote = "",
OPTannote = "",
URL = {http://www.labri.fr/~ramet/restricted/renpar8.ps},
KEYWORDS = "Overlap",
ABSTRACT = {Le cadre g\'en\'eral de ce travail concerne les machines parall\`eles M.I.M.D. \`a m\'emoire distribu\'ee. Dans ce contexte, il est admis que la bonne d\'emarche de gestion du parall\'elisme passe par l'utilisation syst\'ematique de bibliot\`eques performantes de calcul et de communication avec la possibilit\'e de faire du recouvrement pour masquer, dans le temps, les communications par des calculs. Le th\`eme central de ce travail concerne le recouvrement calcul/communication, avec en particulier le calcul, \'eventuellement adaptatif, de la taille optimale des paquets \`a communiquer.}
}