[mpich-discuss] Parallel jobs

[Gus Correa]

Hi Talla

It is no rocket science, just add them to your mpirun command.
Say:

Old command (whatever you have there):
mpirun -n 16 my_program

Modified command (whatever you had there with "-s all" added):
mpirun -s all -n 16 my_program

I am not even sure if it will work, but it is worth giving it a try,
as all processes seem to be trying to read from stdin,
for which in MPICH2 seems to require the specific flag '-s all'.

***

In a different key,
looking at your messages, and the paths that you refer to,
it seems to me that you may be mixing MPICH2 and OpenMPI.
At least the test program you mention this:

 >         I tested the cluster using the following command:
 >         /opt/openmpi/bin/mpirun -nolocal np 16 -machinefile machines
 >         /opt/mpi-test/bin/mpi-ring


In particular, the mpirun commands of each MPI are quite different,
and to some extent so is the hardware support.
This mixing is a very common source of frustration.

Both MPICH2 and OpenMPI follow the MPI standard.
However, they are different beasts,
and mixing them will not work at all.

Stick to one of them only, for compilation (say mpicc or mpif90)
and to run the program (mpirun/mpiexec).
In doubt, use full path names for all (mpicc, mpif90, mpirun, etc).
This mailing list is the support list for MPICH2:
http://www.mcs.anl.gov/research/projects/mpich2/

For OpenMPI you need to look elsewhere:
http://www.open-mpi.org/

You may want to check first how your (ABINIT) computational Chemistry
program was compiled.
Computational Chemistry is *not* my league, I have no idea of what the
heck ABINIT does, probably some many-body Schroedinger equation, and I 
forgot Quantum Mechanics a while ago.
Don't ask me about it.

If your ABINIT came pre-compiled you need to stick with
whatever flavor (and even version) of MPI that they used.
If you compiled it yourself, you need to stick to the MPI flavor and
version associated to the *same* MPI compiler wrapper you used to 
compile it.
I.e. mpicc or mpif90 *and* mpirun/mpiexec must be from the same MPI 
flavor and version.
Take a look at the Makefiles, perhaps at the configure scripts,
they may give a hint.

Good luck.

My two cents,
Gus Correa

Talla wrote:
> Hello Gus,
> Thank you for pointing me to this error, but I am new to Linux world so 
> I am wondering if you have a ready script or commands to add stdin flags 
> to mpirun?
>  
> Thanks
> 
> On Mon, Nov 29, 2010 at 7:34 PM, Gus Correa <gus at ldeo.columbia.edu 
> <mailto:gus at ldeo.columbia.edu>> wrote:
> 
>     Hi Talla
> 
>     It sounds like all processes are trying to read from stdin, right?
>     Stdin/stdout are not guaranteed to be available to all processes.
> 
>     Have you tried to add these flags in your mpiexec/mpirun
>     command line:  "-s all" ?
>     Check the meaning with "man mpiexec".
> 
>     My two cents,
>     Gus Correa
> 
>     Talla wrote:
> 
>         Hello,
> 
>          this issue is consuming all my time with no luck.  I can submit
>         any job without any error message *BUT ONLY* when I am using
>         *one* CPU, when I am using more than one CPU, I got the error
>         message : " *forrtl: severe (24): end-of-file during read, unit
>         5, file stdin* ". I got this error line repeated as many as the
>         number of CPU's I am using.
> 
>         Which mean that the other nodes are not doing any job here. So
>         if you can help me to link the other nodes so I can take
>         advantage of all the nodes.
>         I have 8 PC and each one has 2 CPU (in total I have 16 CPU).
> 
>         I tested the cluster using the following command:
>         /opt/openmpi/bin/mpirun -nolocal np 16 -machinefile machines
>         /opt/mpi-test/bin/mpi-ring
> 
>         and all the nodes can send and receive data like a charm.
> 
>         Just to mention that I have Rocks Clustering software with open
>         centOS.
> 
>         The code I am using is called ABINIT and it is impeded as a
>         plugin in a visual software called MAPS.
> 
>         Your help is really appreciated.
> 
> 
> 
> 
>         ------------------------------------------------------------------------
> 
>         _______________________________________________
>         mpich-discuss mailing list
>         mpich-discuss at mcs.anl.gov <mailto:mpich-discuss at mcs.anl.gov>
>         https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 
> 
>     _______________________________________________
>     mpich-discuss mailing list
>     mpich-discuss at mcs.anl.gov <mailto:mpich-discuss at mcs.anl.gov>
>     https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> mpich-discuss mailing list
> mpich-discuss at mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss