Hi Prakashan,
I believe it might be something from PE setting. Could you try this:
Change this parameter in the 'orte' parallel environment from:
> job_is_first_task TRUE
to:
> job_is_first_task FALSE
If you have this set to true, it would take away an available slot in
your job, so it might prevent an SGE 'task' from launching to one of
your SGE nodes.
Korambath, Prakashan wrote:
Hi,
I just compiled OpenMPI version 1.2.5 with the option
./configure --prefix=/u/local/mpi/openmpi/1.2.5
--with-openib=/usr/local --enable-static --disable-shared CC=icc
CXX=icpc F77=ifort FC=ifort --with-sge
on a X86_64 machine with Infiniband Interconnect and OFED software and
CentOS 5 OS
Everything works fine on command line job submission, but when I submit
through SGE 6.1U3 I am getting following error
error: executing task of job 23081 failed:
[n99:01442] ERROR: A daemon on node n99 failed to start as expected.
[n99:01442] ERROR: There may be more information available from
[n99:01442] ERROR: the 'qstat -t' command on the Grid Engine tasks.
[n99:01442] ERROR: If the problem persists, please restart the
[n99:01442] ERROR: Grid Engine PE job
[n99:01442] ERROR: The daemon exited unexpectedly with status 1.
In my command script for SGE I have
#$ -pe orte 2
/u/local/mpi/openmpi/1.2.5/bin/mpiexec -n 2 -machinefile $TMPDIR/nodefile \
/u/home2/ppk/MPI/C/executablename >& output
n99:/work/23081.1.campus.q {1002}$ cat nodefile
n99 slots=1
n15 slots=1
n99:/work/23081.1.campus.q {1003}$ qconf -sp orte
pe_name orte
slots 360
user_lists NONE
xuser_lists NONE
start_proc_args /bin/true
stop_proc_args /bin/true
allocation_rule $round_robin
control_slaves TRUE
job_is_first_task TRUE
urgency_slots min
I am combing through the archives to look for similar errors. I have
seen some of it, but no satisfactory answer. Anyone knows why?
i02:/u/local/mpi/openmpi/1.2.5/bin {1049}$ ./ompi_info | grep tm
MCA memory: ptmalloc2 (MCA v1.0, API v1.0, Component v1.2.5)
I also tried pre-relese 1.2.6rc3 same results.
Prakashan
------------------------------------------------------------------------
_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
--
- Pak Lui
pak....@sun.com
