Hi,
I am not sure if I understand your question right. If you are asking
whether single MPI_Reduce call is faster than single MPI_Send call
assuming all other parameters like nodes involved in call, amount of data
are constant then I doubt it. There are many factors which might be
important here.
Few I can think of are:
1. Is the resolution of timer you are using good enough for measuring such
small time interval (remember preferably timing interval should be greater
than 100 * timer resolution * timer overhead)
2. Network load might be dynamic. Make sure you take average of some
significant number of readings.
3. If your MPI_Send is blocking for some reason, it will not
complete before matching receive has been posted.
You might want to look into these factors.
cheers,
Manish Chablani
------------------------------------------------------
Graduate Student, CS Department, Indiana University.
http://www.cs.indiana.edu/~mchablan
LAM/MPI Developer
Make today a LAM/MPI day !!!
http://www.lam-mpi.org
------------------------------------------------------
On Tue, 6 May 2003, choy hau yan wrote:
> I have simple question for using mpi_reduce and
> mpi_send/recv.
>
> I found that the time for using mpi_reduce is shorter
> than mpi_send/recv.
> Am I right? Why? Hope that somebody can answer this. thanks.
>
> __________________________________
> Do you Yahoo!?
> The New Yahoo! Search - Faster. Easier. Bingo.
> http://search.yahoo.com
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
|
