torque+mpiexec+mvapich = strange behavior

[Anton Starikov]

Thank you very much for fast fixing.
It works perfect now!

Anton.


Pete Wyckoff wrote:
> A.Starikov at utwente.nl wrote on Mon, 06 Dec 2004 16:10 +0100:
> 
>>I thought the same, but I've add printf after line
>>mport_fd = socket(PF_INET, SOCK_STREAM, 0);
>>in  prepare_ib_startup_port, and this is passed twice for
>>mpiexec -np 1 bla-bla-bla
> 
> 
> That's actually just an artifact of the interaction of stdio redirection
> and pbs_demux.  Add a fflush(stdout) after the printf to make it go
> away.
> 
> Here's another way to cause your IB problem (in bash):
> 
>     exec 0<&-
>     ./mpiexec -np 1 hello
> 
> Apparently recent torque versions just do close(0).  I think traditionally
> PBS would connect the job script or script filename to stdin.
> 
> Regardless, I just tossed the fix into CVS along with a couple
> other inconsiquential changes.  Thanks much for finding this problem
> and leading me towards the solution!
> 
> 		-- Pete
> 
>