Abaqus Batch

Submitting Abaqus Jobs to Batch System: `abaqus_batch`

A script is available for submitting Abaqus 6.6 jobs to the PBS batch queues on the SGI/Altix system at NCSA,
or PBS batch queues on EMt64/linux cluster at NCSA. The script is at /usr/apps/csm/scripts/abaqus_batch.

When the script is invoked with the option --help, it prints a list of its options and exits:

  co-login1% /usr/apps/csm/scripts/abaqus_batch  --help
   
  ABAQUS 6.6-EF1 --- SGI/Altix

  usage: abaqus_batch job_name [options]

  This script submits an Abaqus/Standard job to the batch system.  
  options:
    --help            # display options and exit
    --dryrun          # Check arguments and exit

    --input file      # input file (default is job_name.inp)
    --user_sub file   # compile and link user supplied Fortran subroutines
    --lib library     # link with the given library (or libraries)
    --datacheck       # datachecking (no runs) 
    --double          # double precision executable for Abaqus/Explicit
    --restart file    # use restart file from earlier Abaqus run
    --globalmodel file# global model's results file
    --datafile file   # reads input from external datafile
    --datafile2 file2.tar   # transferes and untars file.tar archive to batch host.
                     #(file.tar should archive all nonstandard external files)


    --fil file        # reads data from previous result file

    --wall time       # wall clock time in minutes, for batch; default is 60 mins
    --mem mem         # memory in Mbytes; default is 500 Mb
    --np np           # number of processors; default is 1

    --no_copy_back    # suppress copying of output files to submit directory;
                      # (output files can still be retrieved from UniTree)
    --mail            # to receive email after job completes
    --proj proj_name  # to charge SU-s to a different then default project (in case of multiple projects) 
    --thread_domain   # Explicit only, Copper only, use threaded  domain decomposition solver
    --mpi_domain      # Explicit only, Copper and Tungsten, use MPI domain decomposition solver 
  cu12%

In normal use, the script prints a summary of requested options and submits a batch job. The
batch job runs Abaqus/Standard or Abaqus/Explicit , creates a gzipped tar file with the user's input files and any
Abaqus output files, and then uploads the tar file to the mass storage system. In addition, all
Abaqus output files are copied to the directory where abaqus_batch was invoked. Email is
sent to the user if the batch job fails.

Stdout and stderr from the execution of Abaqus/Standard(Explicit) are redirected to a single file. The name of the file has the form jobname.log, where jobname is the job name that appears on the abaqus_batch command line.

A typical invocation is shown below with an input deck and a user subroutine, "llq" or ("bjobs" on tun/linux) command is
is excuted to check if job is queued.

  co-login1% ls
  t2.inp mysub.f
  co-login1% abaqus_batch t2 --wall 20 --mem 1500 --user_sub mysub.f

  ABAQUS 6.6-EF1  --- SGI/Altix 

  SUMMARY OF COMMAND LINE ARGUMENTS, DEFAULTS
  job name:       t2
  input file: /u/ncsa/skoric/abq/t2.inp
  number of processors: 1
  submitting a job to the batch queuing system
  estimated cpu time: 20 minute(s)
  estimated memory usage: 1500 Mbyte(s)
  compiling and linking Fortran subroutines in file /u/ncsa/skoric/abq/mysub.f

  
  output files will be copied back to the submit directory
   (a tar file of the output files will also be found on UniTree 
    in your UniTree directory ~/abaqus_batch/)

For complete information, see
http://www.ncsa.uiuc.edu/UserInfo/Resources/Software/CSM/abaqus_batch.html

To check the status of your batch jobs type qstat -u your_login on SGI/Altix or bjobs on Xeon/Linux cluster.

Please keep in mind that batch jobs are killed by the batch system, if they use more wall
time than they request. Check the job reports mailed by the batch system to find out how
much wall clock time and memory your jobs use, and plan walll clock time and memory requests for
future jobs accordingly. Use datacheck option in interactive run, or --datacheck option in batch to
check for necessary memory requirements for your run.

All the files are regularly purged from scratch, and it might happen that your large input files
are purged even before your job starts executing from batch (while it is still pending in
batch). To prevent this, submit your large inputs from your home directory and use
--no_copy_back option to suppress copying back large outputs to your home directory. Your
tarred outputs can be still found on UniTree in your home UniTree directory ~/abaqus_batch/.

The tar file that's uploaded to the mass storage system at the end of each job is placed in~/abaqus_batch in the user's mass storage system home directory. Uploaded tar files have
names of the form jobid.tar.gz, where jobid is the 4(or 5)digit job-id assigned by the batch
system.

--fil result_file_name If your job reads data from a previous results file (*.fil), then this option just copies *.fil file to the directory on the execution host where your job will be running. This is not fil option from interactive execution procedure. Use this option in batch only if you are reading results files from previous runs.

--datafile data_file_name Similar to --fil, this option copy to the batch host an external file that your job reads from (things like amplitude, node, or element data).

--globalmodel global_model_results_file This option specifies the name of the global model's results file from which the results are to be interpolated to drive a submodel analysis. This option is required whenever the *SUBMODEL or *BOUNDARY SUBMODEL option is in the input file.

--datafile2 file.tar Use this option if there are many non standard Abaqus files that need to be copied to the batch host.
You must tar all non standard files into an archive first. Please see FAQ #13.

This script does not have any options for interactive use of Abaqus.

abaqus_batch

Submitting Abaqus Jobs to Batch System: abaqus_batch

Submitting Abaqus Jobs to Batch System: `abaqus_batch`