This is an old revision of the document!
The interactive partition
int is meant for interactive jobs, i.e.,
- jobs that require constant user interaction (like Mathematica or other GUI oriented packages),
- test sessions with many and longer runs, beyond what we can tolerate on the frontends.
It is not meant for jobs running constantly for hours, even you have some interaction with them at the beginning.
The special property of this queue is that the participating nodes have no slot limit. They will take jobs until their load crosses a specified threshold. This accommodates the fact that interactive job cores are not used constantly.
Submitting interactive jobs works as follows (sessions started without srun will be automatically killed):
For an interactive bash session:
srun -p int -n 20 -N 1 --pty bash
For an interactive bash session with X-Windows (GUI) support:
srun --x11 -n 20 -N 1 -p int --pty bash
For an interactive session with MPI support:
srun -p int -n 80 -N 4 --ntasks-per-node=20 --pty bash
You can also use these commands with a standard queue, for example to set some parameters interactively at the beginning of a long running job.
The default walltime in this queue is 1 hour, you can extend the walltime to a maximum of 12 hours, by using “-t”.
NEED REVIEW FOR SLURM:
Note that only in the int queue you are allowed to use more cores than you asked for. However, mpirun will always start the number of processes specified with srun. While you can also use other means (like mpiexec) to start MPI jobs in the int queue, processes running on a node not participating in your interactive job will be automatically killed.