[ExM Users] Submitting tasks on Raven

[Justin M Wozniak]

This definitely works now- we were able to launch tasks on 6 GPUs on Raven.

On 04/25/2013 09:40 AM, Michael Wilde wrote:
> Can someone help Scott with this?
>
> Scott, I think #PBS -m is the email notification flag, not memory.
>
> Its complaining that your aprun command is asking for more resources
> than the qsub command requested for the job.
>
> Check that the aprun -n -N -d do not exceed mppwidth etc from pns
>
> Test your aprun args with a qsub -I asking for the correct numnber of
> nodes and cores
>
> On 4/24/13, Scott Krieder <skrieder at iit.edu> wrote:
>> Hi All,
>>
>> I'm trying to run noop.tcl with turbine on Raven. I keep getting a memory
>> error:
>> apsched: claim exceeds reservation's memory
>>
>> I tried a few different values(100, 100M) for
>> #PBS -m
>> but I keep getting the error.
>>
>> Is there a way to let the PBS job take as much memory as it needs?
>>
>> Thanks,
>> Scott
>>
>> =====aprun.sh script that I'm running=====
>> # USAGE: qsub aprun.sh
>>
>> # The user should copy and edit the parameters throughout this script
>> # marked USER:
>>
>> # USER: Directory available from compute nodes:
>> USER_WORK=/ufs/home/users/p01684
>>
>> # USER: (optional) Change the qstat name
>> #PBS -N turbine
>> # USER: Set the job size
>> #PBS -l mppwidth=1,mppnppn=3,mppdepth=1
>> # USER: Set the wall time
>> #PBS -l walltime=10:00
>> # USER: (optional) Redirect output from its default location ($PWD)
>> #PBS -o /ufs/home/users/p01684/pbs.out
>>
>> #PBS -j oe
>> #PBS -m n
>>
>> # USER: Set configuration of Turbine processes
>> export TURBINE_ENGINES=1
>> export ADLB_SERVERS=1
>>
>> echo "Turbine: aprun.sh"
>> date "+%m/%d/%Y %I:%M%p"
>> echo
>>
>> # Be sure we are in an accessible directory
>> cd $PBS_O_WORKDIR
>>
>> set -x
>> # USER: Set Turbine installation path
>> export TURBINE_HOME=${USER_WORK}/Public/sfw/turbine
>> # USER: Select program name
>> # PROGRAM=${USER_WORK}/adlb-data.tcl
>> PROGRAM=${TURBINE_HOME}/test/noop.tcl
>>
>> source ${TURBINE_HOME}/scripts/turbine-config.sh
>> if [[ ${?} != 0 ]]
>> then
>>    echo "turbine: configuration error!"
>>    exit 1
>> fi
>>
>> # Send environment variables to PBS job:
>> #PBS -v TURBINE_ENGINES ADLB_SERVERS TURBINE_HOME
>> # USER: Set aprun parameters to agree with PBS -l settings
>> # aprun -n 1 -N 1 -cc none -d 1 ${TCLSH} ${PROGRAM}
>> aprun -n 3 -N 3 -cc none -d 1 ${TURBINE_HOME}/bin/turbine ${PROGRAM}
>>
>> --
>> Scott J. Krieder
>> C: 419-685-0410
>> E: skrieder at iit.edu
>> http://datasys.cs.iit.edu/~skrieder/
>>


-- 
Justin M Wozniak