[mpich-discuss] algorithm used in braodcast operation
eya zaibi
zaibi.eya at hotmail.fr
Wed Jan 26 05:32:25 CST 2011
Hi everybody,
I am sorry because i found troubles in using mailing list and that's why i send my question by mail.
I am working in performances prediction of collective routines and i am using algorithmes used in the implementation of MPICH2.
The problem is i didn't the paking of datatype (that was homogeneous ). And i use benchmarking of point to point routines in algorithme of broadcast i.e. that i conserve the algorithme of bcast , i use the results of benchmarking of send/recv but i eliminate (the paking datatype ).
I do some tests but the time estimated is much less than reality.
My question is : is what this work is sufficient to prdict the performances of collective routines on heterogeneous environnement (cluster) and what do you suggest me to improve my work?
I will be gratefull for your help and thank you.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20110126/e9e97fd8/attachment.htm>
More information about the mpich-discuss
mailing list