I believe the btl must include self - can't just be openib. So you need -mca btl self,openib, otherwise the proc cannot talk to itself.


On Oct 12, 2011, at 12:46 PM, Bhargava Ramu Kavati wrote:

Hi,
I am using OpenMPI version 1.4.3 on CentOS5.4 machines (connected back to back using Infiniband HW)
I am trying to run example apps in OpenMPI using the below command.

"mpirun --prefix /usr/local/ -np 2 --mca btl openib --mca btl_openib_cpc_include rdmacm -hostfile tmp_host_file ring_c

Here tmp_host_file contains IP addresses of two hosts.  After running this command, I see the error as

"At least one pair of MPI processes are unable to reach each other for
MPI communications.  This means that no Open MPI device has indicated
that it can be used to communicate between these processes.  This is
an error; Open MPI requires that all MPI processes be able to reach
each other.
"
This clearly indicates that there is an option missing the command due to which MPI processes are not able to reach each other.  Can you please let me know
1) how do we resolve this
2) IP is the only way for MPI processes to reach each other (i.e. it'll go via IPoIB in infiniband case) ?

Thanks & Regards,
Ramu
_______________________________________________
devel mailing list
devel@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel