Interactive sessions¶

An interactive session gives you a shell on a compute node, for example to try out commands, compile code or run IPython, before you write a batch job.

Choose a partition¶

Use the interactive partitions. Jobs there start with higher priority than batch jobs.

Cluster	Partition	Nodes	Time limit	Use
Phoebe	`cpu_int`	`n[1-20]`	20 days 10 h	CPU work
Phoebe	`gpu_int`	`gpu[1-2]`	20 days 10 h	GPU work
Phoebe	`small_int`	`s[1-3]`	7 days 7 h	light work (8 cores, 64 GB per node)
Koios	`cpu_int`	`n[11-12]`	9 days 1 h	CPU work

Start a screen session¶

An interactive session ends when you disconnect. Log in to the front-end node and start a screen session first, so the session survives a dropped connection. Here we name it "session007":

[user@login1 ~]$ screen -S session007

Detach with Ctrl+A, then d (two separate key presses; see screen sessions). To come back later, log in again and reattach:

[user@login1 ~]$ screen -r session007

Request resources from Slurm¶

Example 1: 16 CPUs on a CPU node¶

[user@login1 ~]$ srun --partition=cpu_int --job-name "interactive" --cpus-per-task=16 --mem=32G --time=08:00:00 --pty /bin/bash
srun: job 1962744 queued and waiting for resources
srun: job 1962744 has been allocated resources
[user@n11 ~]$

The prompt changed from user@login1 to user@n11: you are now on compute node n11.

Example 2: 2 GPUs and 16 CPUs on a GPU node¶

srun --partition=gpu_int --job-name "interactive" --gres=gpu:a100:2 --cpus-per-task=16 --mem=128G --time=24:00:00 --pty /bin/bash

See GPU jobs for how many CPUs to request per GPU.

When you are done¶

The session ends when its time limit is reached or when you leave the shell with exit. Exit as soon as you are done, so the resources are free for others.