Batch Jobs   (LoadLeveler)

Batch jobs are submitted to the LoadLeveler system via

llsubmit <jobscript>

You just have to specify memory consumption and number of CPUs or cores.
Below you find some examples LoadLeveler scripts for serial and parallel jobs.

For an in depth explanation of selected advanced job command file keywords refer to these explanations.

For a complete description see the LoadLeveler manual and documentation on the Parallel Environment Library:  IBM Manuals

Please note:

• DKRZs Power6 system is run in SMT mode which allows to execute two parallel threads on each physical core!
  Therefore, most programs perform best with 64 tasks or threads per node. For programs with reasonable scaling behaviour this should be up to 20% faster than using only one thread per core (ST mode, 32 tasks or threads per node). In some cases, however, it may be more efficient to use only 32-way parallelism per node. You may wish to try both versions. Examples for both, SMT and ST are given below
  

• The default memory assigned to a serial job 1 GB, the maximum value is 8GB.
  In order to allow as many serial jobs as possible to run simultaneously, we ask you to check your memory requirements in serial jobs. You may use  getrusage (man getrusage) to verify how much memory your job has needed. So you can adjust your memory request of the next serial job accordingly. 
• Even for parallel jobs the required memory should not be set greater than really needed since the cluster is equipped with 80 nodes having 96GB memory available and 160 nodes with only 48GB. Hence, your job might idle for longer time if resources are overestimated.
  
  

Working with several accounts

#@ account_no = <Project>            (e.g.:  #@ account_no = ab0123)