[MPICH] mpich1 X mpich2 rsh troubles

[William Gropp]

At 08:02 AM 3/2/2006, Márcio Ricardo Pivello wrote:
>Hi
>
>I'm starting to work with mpich now, and I'm having some problems when 
>running an application. I'm using mpich2-1.0.2p1 to compile a code which 
>uses petsc 2.3.0, hypre 1.9.0b and ParMetis 3.1. The code has f77 and f90 
>files. Compilation runs without any problem,  but when I try to run it i 
>have the same error, no matters what machines are used in the cluster:
>
>-- Fatal error in MPI_Comm_size: Invalid communicator, error stack:
>MPI_Comm_size(110): MPI_Comm_size(comm=0x1, size=0xbfffe340) failed
>MPI_Comm_size(69): Invalid communicator
>rank 0 in job 4  no43_33215   caused collective abort of all ranks
>   exit status of rank 0: return code 13

This looks like the file was compiled with a a version of mpi.h or mpif.h, 
or an MPI module, that belongs to a different MPI implementation.  Check 
the compilation steps to make sure that the MPICH2 include paths are 
used.  Also try using MPICH2 1.0.3, which is the current release.

Bill

>in this case, no43 is the hostname of the machine. It happens since the 
>first host listed in the hosts file, and the process stops immediately.
>I'm using the following sequence to launch the application:
>
>mpdboot -np 4 -h hosts --rsh=rsh
>rsh <some_machine_listed_in_file_hosts>
>mpdrun -np 4 ../bin/linux-gnu-opt/SolverGP.x -ksp_gmres_restart 90 
>-ksp_rtol 0.000001 -log_info -ksp_monitor -log_summary >& run-2p-b.log &
>
>What is the problem here?
>
>
>Thanks in Advance
>
>-Márcio Ricardo Pivello
>
>PS.: I tried to run this code with mpi1 1.2.6, but the code I'm trying to 
>compile needs the library libmpichf90.a, which I didn't find in mpich 1. 
>Is there any alternative for this library, or could I add this library to 
>mpich1 libraries?
>
>
>
>Márcio Ricardo Pivello
>Mechanical Engineer, MSc.
>LTCM and CFD Lab
>Mechanical Engineering School
>Federal University of Uberlândia
>Uberlândia - MG - Brazil

William Gropp
http://www.mcs.anl.gov/~gropp