Changes between Initial Version and Version 1 of Ticket #181


Ignore:
Timestamp:
08/05/08 19:15:20 (17 years ago)
Author:
arango
Comment:

Oh yes, this is a very good idea.

Legend:

Unmodified
Added
Removed
Modified

  • Ticket #181

    • Property Resolution → Fixed
    • Property Status new → closed
    • Property Summary swan coupling mpi finalize → SWAN coupling mpi_finalize

  • Ticket #181 – Description

    initial v1  
    1 for the swan coupling, the call mpi_finalize is now located at the bottom of 
    2 Master/mct_coupler.h
     1for the '''SWAN''' coupling, the call '''mpi_finalize''' is now located at the bottom of '''Master/mct_coupler.h'''
    32
    43Here is the problem:
    5 It is not clear which model will get there first.  Typically this will be roms. However, swan needs time to collect the contents of the individual output files and conglomerate them into single output files (don't ask me why swan does this, it is just the way it is). But if roms goes thru it's finalize step, then calls mpi_finalize, well that is bad news for the swan because it may not have finished pulling together the output.
     4
     5It is not clear which model will get there first.  Typically this will be '''ROMS'''. However, '''SWAN''' needs time to collect the contents of the individual output files and conglomerate them into single output files (don't ask me why '''SWAN''' does this, it is just the way it is). But if '''ROMS''' goes thru it's finalize step, then calls '''mpi_finalize''', well that is bad news for the '''SWAN''' because it may not have finished pulling together the output.
     6
    67Suggested Solution:
    7 place an mpi_barrier in mct_coupler.F before the call to mpi_finalize so that all processors from all models need to check in before finalize occurs.
     8
     9Place an '''mpi_barrier''' in '''mct_coupler.h''' before the call to '''mpi_finalize''' so that all processors from all models need to check in before finalize occurs.