On Wed, Nov 16, 2011 at 03:52:05PM +0000, Kharche, Sanjay wrote:
> Dear All
> I am sure this issue has been discussed before on the forum, but I will appreciate your comments.
> I have a package that tries to do parallel file output using MPI_File_Write:
> /* Write to file. */
> mpi_errno = MPI_File_write(file, New, 1, sourceType, MPI_STATUS_IGNORE);
> With an increasing number of processors, I see that this causes the file output to take longer. Can someone suggest a solution?
Think a bit about what adding more processors will do. Each MPI
process will write 1 sourceType to the file. More processors will
write more data.
I don't know how your program creates sourceType, nor do i know the
file view (if any) it has placed on the output, so maybe you will need
to show more code.
I hope you are setting a file view here, or each processor will end up
writing the same data to the same location in the file. If you
duplicate the work identically to N processors then yeah, you will
take N times longer.
Mathematics and Computer Science Division
Argonne National Lab, IL USA