[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] MPI job submission script configuration



I believe the manual section “2.9.3 Parallel Jobs with Separate Requirements” may help if you can explicitly specify your machines. If you can’t list machine names, you may be able to get away with doing something like this (since each machine only has one slot with ID 1):

 

Requirements = (SlotID == 1)

 

-Mick

 

 

 

 

From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Kunal Rao
Sent: Thursday, December 08, 2011 9:09 AM
To: condor-users@xxxxxxxxxxx
Subject: Re: [Condor-users] MPI job submission script configuration

 

Any suggestions on the below .. ?

Thanks & Regards,
Kunal

On Mon, Dec 5, 2011 at 1:27 PM, Kunal Rao <kunalgrao@xxxxxxxxx> wrote:

Hi,

  I have setup condor pool with 2 dedicated resources (for MPI jobs), one of them with dedicated scheduler. I am running MPI jobs (MPICH2) with say 2 processes, I want these to go on 2 different machines (the one's which are configured as dedicated resources)

  What I am seeing is that condor creates 2 processes and puts them on the same machine. How do I specify in the job submission script for it to start these 2 processes on different machines ?

  My current job submission script is as follows :

  ---------
  universe = parallel
  executable = /var/spool/condor/etc/examples/mp2script
  arguments = /home/kunal/condor/examples/hello_mpi
  machine_count = 2
  +WantIOProxy = True
  Output = hello_mpi.out
  error = hello_mpi.err
  Log = hello_mpi.log
  should_transfer_files = yes
  when_to_transfer_output = on_exit
  +ParallelShutdownPolicy = "WAIT_FOR_ALL"
  transfer_input_files = /home/kunal/condor/examples/hello_mpi
  queue
  -------

  I want to specify something like: start process 1 on machine1 and process 2 on machine2. How do I specify that ?

Thanks & Regards,
Kunal