Submitting Jobs via condor from a 64bit machine to a 32bit cluster

When submitting jobs via condor_submit, condor will by default transfers the executable files from your local machine. This obviously is a problem if the machines are based on different architectures. If you are unaware of condors default behaviour, this can cause some confusion (especially since condor apparently does not report any errors in this case). For example the follwing condor submission script, will send your local binary 'hostname' file to the grid resource rtgrid1.its.uiowa.edu:

  universe = globus  
  globusscheduler = rtgrid1.its.uiowa.edu/jobmanager-condor   
  executable = /bin/hostname  
  output = why_error_21.out   
  error = why_error_21.err   
  log = why_error_21.log   
  notification = never  
  queue

If you want to run the hostname program on the remote resource you will have to add the following lines to your condor submit script before the 'queue' command:

  • transfer_executables = False
  • copy_to_spool = False

The updated condor submit script would look like this:

  universe = globus  
  globusscheduler = rtgrid1.its.uiowa.edu/jobmanager-condor   
  executable = /bin/hostname  
  output = why_error_21.out   
  error = why_error_21.err   
  log = why_error_21.log   
  notification = never  
  transfer_executables = False
  copy_to_spool = False
  queue

Solution:

This issue has been resolved as of Condor 6.8, if you are experiencing this problem, please upgrade your Condor installation.