[mvapich-discuss] troubles in running MPI job over RoCE, with mvapich2-1.6 shipped with OFED1.5.3.2

[Jonathan Perkins]

On Thu, Feb 21, 2013 at 09:06:05PM +0530, Devesh Sharma wrote:
> Hi Jonathan and Devendar, Thanks for a quick response.
> 
> MV2_USE_RoCE-1 is given in section 6.12 of mvapich2-1.8.1 user guide. I
> have taken this from there.
> I have installed xterm, it was not there. and also changed the parameter
> names as suggested. I am hitting following output:
> 
> [root at neo01 IMB-3.2]# /usr/mpi/gcc/mvapich2-1.6/bin/mpirun_rsh -ssh -debug
> -np 2 MV2_USE_RDMAOE=1 MV2_USE_RDMA_CM=1 -hostfile /opt/Work/hostfile
> /bin/hostname
> Without hostfile option, hostnames must be specified on command line.
> usage: mpirun_rsh [-v] [-sg group] [-rsh|-ssh] [-debug] -[tv] [-xterm]
> [-show] [-legacy] -np N(-hostfile hfile | h1 h2 ... hN) a.out args |
> -config configfile (-hostfile hfile | h1 h2 ... hN)]

It looks like the order of your parameters are causing the problem.
Move the environment variables to after your mpirun_rsh options.

Example:
/usr/mpi/gcc/mvapich2-1.6/bin/mpirun_rsh -ssh -debug -np 2 -hostfile /opt/Work/hostfile MV2_USE_RDMAOE=1 MV2_USE_RDMA_CM=1 /bin/hostname

Please note that mvapich2-1.6 is old and no longer supported.  You
should upgrade to mvapich2-1.8.1.

> 
> One basic doubt I have. I remember long time back as a part of MPI setup I
> use to create a password file in the home directory of user. In this file
> we use to specify a passcode. Is it still a requirement?

No.

-- 
Jonathan Perkins
http://www.cse.ohio-state.edu/~perkinjo