Putting your Condor^® pool on the Grid with the Globus Toolkit^®

You can submit Globus jobs that will run on a Condor pool. You can do this by configuring Globus to use the Condor jobmanager, which is provided with the Globus Toolkit. The Globus manuals have some description of how to do this, but to simplify your life, we will give some more detailed information. In order to follow these instructions, we assume two things:

• You have already installed Globus 2.0 or 2.2, and gotten it running correctly.
• You have already installed Condor and gotten it running correctly.

Assuming you're ready to go, you'll need to follow one of two courses of action, one for a Condor pool that has just one operating system and architecture (a homogenous pool), and one for a Condor pool that has multiple operating systems or architectures (a heterogeneous pool).

1. Make sure GLOBUS_LOCATION and/or GPT_LOCATION are set appropriately.
2. Run the following command:
   

       $GLOBUS_LOCATION/setup/globus/setup-globus-gram-job-manager -type=condor
   

3. Edit $GLOBUS_LOCATION/etc/grid-services/jobmanager-condor to add the parameters -condor-arch and -condor-os, which specify what ARCH/OPSYS to use. The results should look something like this (everything on a single line), except that you will change the pathnames to be correct for your system:
   

   stderr_log,local_cred - /scratch/globus/libexec/globus-job-manager globus-job-manager -conf /scratch/globus/etc/globus-job-manager.conf -type condor -rdn jobmanager-condor -machine-type unknown -condor-arch INTEL -condor-os LINUX
   

4. Run the following command:
   

       $GLOBUS_LOCATION/setup/globus/setup-globus-gram-reporter -type=condor
   

For a heterogeneous pool, things are a little more complicated than the homogeneous case. This is because you need to setup a jobmanager for each platform in your pool that you want to be able to submit to:

1. Make sure GLOBUS_LOCATION and/or GPT_LOCATION are set appropriately.
2. Change the Globus etc directory:
   

       cd $GLOBUS_LOCATION/etc
   

   Now, perform the following steps once for each platform in your pool:

3. Make a copy of the globus-job-manager.conf file (the new file will be specific to this particular platform):
   

       cp globus-job-manager.conf globus-job-manager-condor-INTEL-LINUX.conf
   

4. Edit the -globus-host-cputype/manufacturer/osname/osversion options in the new conf file to match those of the platform.
5. Make a copy of the grid-services/jobmanager file (the new file will be specific to this particular platform). The name of this file will be the name that user will use to submit jobs to this platform:
   

       cp grid-services/jobmanager grid-services/jobmanager-condor-INTEL-LINUX
   

6. Edit the new file in grid-services to change the -conf, -type, and -rdn paremeters and add the -condor-arch and -condor-os parameters (which specify which ARCH/OPSYS to use). The end result should look something like this (all on one line), except that you will change the pathnames to be correct for your system::
   

   stderr_log,local_cred - /scratch/globus/libexec/globus-job-manager globus-job-manager -conf /scratch/globus/etc/globus-job-manager-INTEL-LINUX.conf -type condor -rdn jobmanager-condor-INTEL-LINUX -machine-type unknown -condor-arch INTEL -condor-os LINUX