[mvapich-discuss] can't run mvapich MPI job in background

Benjamin Rutt rutt at bmi.osu.edu
Wed Apr 19 12:29:25 EDT 2006


Sayantan Sur <surs at cse.ohio-state.edu> writes:

> I am wondering why you get a message "Stopped"? 

I was using 'set -b' in bash which tells me any status change in a
background job immediately.

> Ideally, the process is just backgrounded ... not stopped. Did you
> by chance hit Ctrl-Z on your keyboard?

No, I didn't.

However, I found 2 workarounds, I think.  I kept playing with it and
got this output:

    [rutt at node23 mpitest]$ mpirun_rsh -rsh -np 2 node23 node24 ./cpi &
    [1] 3728
    [rutt at node23 mpitest]$ 

    [1]+  Stopped                 mpirun_rsh -rsh -np 2 node23 node24 ./cpi
    [rutt at node23 mpitest]$ 
    [rutt at node23 mpitest]$ fg
    mpirun_rsh -rsh -np 2 node23 node24 ./cpi
    rcmd: poll (setting up stderr): Interrupted system call
    Signal 2 received.
    done.

I thought maybe it had a problem with stdin/out/err, so I did:

    [rutt at node23 mpitest]$ mpirun_rsh -rsh -np 2 node23 node24 ./cpi < /dev/null &
    [1] 3776
    [rutt at node23 mpitest]$ Process 0 of 2 on node23
    Process 1 of 2 on node24
    pi is approximately 3.1415926544231318, Error is 0.0000000008333387
    wall clock time = 0.000207

    [1]+  Done                    mpirun_rsh -rsh -np 2 node23 node24 ./cpi </dev/null
    [rutt at node23 mpitest]$ 

So redirecting stdin from /dev/null seems to fix it completely.  I
also found that using -ssh instead of -rsh for job startup fixes it
too.  Thanks,
-- 
Benjamin


More information about the mvapich-discuss mailing list