From CMMS

Load Leveler Guide lines

LoadLeveler 3.2 is currently installed only on the IBM p655 Machine.

[edit] Submission Scripts

The following is a sample script, 4_cpu_example.cmd, to submit a 4 processor job to the p655_cmms_4 queue.

      #!/bin/ksh
       
      # @ job_type=parallel
      # @ job_name=4_cpu_example
      # @ error=4_cpu_example.error
      # @ checkpoint=no
      # @ restart=no
      # @ class=p655_cmms_4
      # @ node_usage=shared
      # @ notification=never
      # @ group=[my group]

      /path/to/my/program < 4_cpu_example.inp > 4_cpu_example.log

      # @ queue

This file is submitted to LoadLeveler using llsubmit:

/usr/bin/llsubmit 4_cpu_example.cmd

Please be careful when preparing your script. If you are using the p655 machines your job_type is always parallel. The node_usage variable can have one of two values, shared and not_shared. If you are using a 4 CPU class with not_shared then your job will have complete possession of the node and no other job will be started.

[edit] Gaussian03

If you are using Gaussian03 you can submit your job with /usr/local/bin/gog03. The example input below is for an 8 processor optimization of the water dimer (w2-opt.inp) and the last line is the syntax of gog03 to submit the job to the p655_cmms_8 class.

Note that gaussian inputs must have .inp extension and the extension is left off when running gog03.

      %nproc=8
      %mem=2000mb
      %chk=w2-opt.chk
      # mp2/aug-cc-pV5Z scf=tight maxdisk=30GB opt=z-matrix

      water dimer optimization

      0 1
      O
      H 1 roh1
      H 1 roh2 2 ahoh
      O 1 roo 2 aooh 3 doohh 0
      H 4 roh3 1 ahoo 2 dhooh 0
      H 4 roh3 1 ahoo 2 -dhooh 0
      Variables:
      roo 2.9098
      roh1 0.9605
      roh2 0.9686
      roh3 0.9624
      ahoh 104.3433
      aooh 109.2085
      ahoo 109.8387
      dhooh 122.7631
      Constants:
      doohh 0.

you@cmms23:~/g03-inputs>/usr/local/bin/gog03 -q p655_cmms_8 w2-opt

If you do not specify %nproc, gog03 will add it to your input if you select a parallel queue.

Extreme care must be taken with %mem in your input. Even if you specify all the memory available, your job may not run efficiently and perhaps not even in parallel. Please consult The Gaussian Online Manual before running your job.

[edit] Managing your job

After you have submitted your job, you can check its status with the command llq, which will show the Job Id, owner, date of submission, the class, and node where the job is running. Should you need to cancel a job, use llcancel [Job Id].