Interactive Jobs

The interactive (floating) partition (exclusively associated to the debug QOS) is to be used for code development, testing, and debugging.

Important

Production runs are not permitted in interactive jobs. User accounts are subject to suspension if they are determined to be using the interactive partition and the debug QOS for production computing. In particular, interactive job "chaining" is not allowed. Chaining is defined as using a batch script to submit another batch script.

You can access the different node classes available using the -C <class> flag (see also List of Slurm features on ULHPC nodes), or (better) through the custom helper functions defined for each category of nodes, i.e. si, si-gpu or si-bigmem:

Regular Dual-CPU node

### Quick interative job for the default time
$ si
# salloc -p interactive --qos debug -C batch

### Explicitly ask for a skylake node
$ si -C skylake
# salloc -p interactive --qos debug -C batch -C skylake

### Use 1 full node for 28 tasks
$ si --ntasks-per-node 28
# salloc -p interactive --qos debug -C batch --ntasks-per-node 28

### interactive job for 2 hours
$ si -t 02:00:00
# salloc -p interactive --qos debug -C batch -t 02:00:00

### interactive job on 2 nodes, 1 multithreaded tasks per node
$ si -N 2 --ntasks-per-node 1 -c 4
si -N 2 --ntasks-per-node 1 -c 4
# salloc -p interactive --qos debug -C batch -N 2 --ntasks-per-node 1 -c 4

GPU node

### Quick interative job for the default time
$ si-gpu
# /!\ WARNING: append -G 1 to really reserve a GPU
# salloc -p interactive --qos debug -C gpu -G 1

### (Better) Allocate 1/4 of available CPU cores per GPU to manage
$ si-gpu -G 1 -c 7
$ si-gpu -G 2 -c 14
$ si-gpu -G 4 -c 28

Large-Memory node

### Quick interative job for the default time
$ si-bigmem
# salloc -p interactive --qos debug -C bigmem

### interactive job with 1 multithreaded task per socket available (4 in total)
$ si-bigmem --ntasks-per-node 4 --ntasks-per-socket 1 -c 28
# salloc -p interactive --qos debug -C bigmem --ntasks-per-node 4 --ntasks-per-socket 1 -c 4

### interactive job for 1 task but 512G of memory
$ si-bigmem --mem 512G
# salloc -p interactive --qos debug -C bigmem --mem 512G

If you prefer to rely on the regular srun, the below table proposes the equivalent commands run by the helper scripts si*:

Node Type	Slurm command
regular si [...]	salloc -p interactive --qos debug -C batch [...] salloc -p interactive --qos debug -C batch,broadwell [...] salloc -p interactive --qos debug -C batch,skylake [...]
gpu si-gpu [...]	salloc -p interactive --qos debug -C gpu [-C volta[32]] -G 1 [...]
bigmem si-bigmem [...]	salloc -p interactive --qos debug -C bigmem [...]

Impact of Interactive jobs implementation over a floating partition

We have recently changed the way interactive jobs are served. Since the interactive partition is no longer dedicated but floating above the other partitions, there is NO guarantee to have an interactive job running if the surrounding partition (batch, gpu or bigmem) is full.

However, the backfill scheduling in place together with the partition priority set ensure that interactive jobs will be first served upon resource release.

---

Last update: November 13, 2024