LoadLeveler and mpirun
On Hebb a job or application runs in its own block and takes over the entire block. Each user job can take anywhere from 32 Compute Nodes (the smallest block size), up to taking over the entire system.
On Hebb we are using a job scheduler called '''LoadLeveler''', which directly invokes mpirun.
When you login on Hebb you enter the '''front end node'''. In order to submit your job with LoadLeveler you need to move your code to the cluster wide file system GPFS.
Using a LoadLever submit script you for instance can tell the scheduler how many compute nodes you want to use # @ bg_size = 32 or what class your job belongs to # @ class = large. Then, the scheduler assigns compute nodes and your job starts when the requested nodes are available.
The available LoadLeveler commands are:
ll llacctval llcancel llclass llq llstatus llsubmit
Appending the '''-H''' flag to any of these will give you details on the command. For a smaller summary, instead add the '''-?''' flag.