[MPICH] problems running a program using MPI

Michael Baxa baxa at uchicago.edu
Fri Feb 22 11:06:30 CST 2008


Hello,

They may very well be elementary issues, but I have two problems that 
may or may not be related.  These pertain to a Fortran 90 program that 
is supposed to use multiple nodes on a Torque (PBS) and Maui job queue 
server

Problem #1: Before I call my program, I boot mpd with the following:  ` 
mpdboot --file=$PBS_NODEFILE `

However, only the first node in $PBS_NODEFILE is booted.  In an 
interactive session, I try to get around this by ssh'ing to the other 
nodes in $PBS_NODEFILE and calling ` mpd -h <node1> -p <node1 port> `.  
After doing this, `mpdtrace` returns all of the nodes in $PBS_NODEFILE.  
I don't know why this isn't done automatically in the first place.

Problem #2: With mpd now booted on all my nodes, I now call my program 
using

`mpirun -machinefile $PBS_NODEFILE -np <number of nodes> <program>`

The program seems to hang the first time it comes to MPI_BCAST, i.e. for 
an array named temps, I have the following

    do i=0,nproc-1
      print *, 'i = ',i
      call MPI_BCAST (temps(i), 1, MPI_DOUBLE_PRECISION, i, 
MPI_COMM_WORLD, ierr)
    enddo

The last thing the program outputs is 'i =            0`

One side note, I put a 'call MPI_BARRIER(MPI_COMM_WORLD,ierr)` before 
this do loop just to see what would happen, and every node froze at that 
line.

I do not know if this is a relevant issue to this forum, but I would 
greatly appreciate any suggestions.

Thank you,

Michael Baxa







More information about the mpich-discuss mailing list