The IT department of the Institut of Physics maintains a collection of commands which simplify the interaction with Slurm. Feature-requests and Bugreports are welcome! Contact IT-Physik.
These commands have to be activated by loading a module.
module load slurm-helper
A small overviw of the available commands will be displayed.
module load slurm-helper 2>/dev/null
List own active and pending jobs with additional information
q #this comment prevents confluence to just display an empty box
List all active and pending jobs with additional information
qu #this comment prevents confluence to just display an empty box
Output in the same fashion as q
Show Node occupation in terminal (basic sview alternative)
qq #this comment prevents confluence to just display an empty box
Show ressource-efficiency (CPU, memory) of RUNNING Jobs
Especially users running parallel MPI or OpenMP (Multithreading) Jobs should make sure that their Jobs are running effectively (= using all the requested ressources).
sperf #this comment prevents confluence to just display an empty box
In this example the requested CPU-Cores are used rather efficiently while most of the requested memory is not used at all, which should be decreased.
Behind the scenes sstat command (from Slurm) is used. DO NOT use this command if you are running a lot of Jobs and DO NOT use this command via watch .
Show ressource-efficiency (CPU, memory) of FINISHED Jobs
seff -d 1 #this comment prevents confluence to just display an empty box
By default only Jobs using more than 1 CPU core and running more than 5 minutes are shown. Use -t to also include shorter Jobs and -1 to also include Jobs with only 1 CPU core.
The command seff has a couple of options which are displayed when running without arguments.
Show partition occupation and user statistics
uu #this comment prevents confluence to just display an empty box
