Reference

Last updated on 2026-10-27 | Edit this page

Snellius places

The course cluster does not have this list. Use the tab in the episode “The file systems”.

Place	Path	Backup	How long it lasts	Quota
Home	`/home/$USER`	Yes, nightly	Project end, plus 15 weeks. Deleted files can be restored for three weeks	200 GiB
Scratch-shared	`/scratch-shared/$USER`	No	14 days	Counted in the 8 TiB scratch quota
Scratch-local	`$TMPDIR`	No	6 days	Counted in the scratch quota
Scratch-node	`$TMPDIR` with `--constraint=scratch-node`	No	Removed when the job ends	No quota
Project	`/projects`	No	The life of the project	On demand
Archive	`/archive/$USER`	Yes	The life of the project	On demand, tape

Commands

BASH

pwd
mkdir -p /scratch-shared/$USER
ls -l
chmod 640 mode-demo.txt
chmod g-r mode-demo.txt
myquota
df -h
du -sh ~

chmod g-r removes group read. A file that started as -rw-r----- becomes -rw-------.

A job directory:

BASH

ls -l "$TMPDIR"
touch "$TMPDIR/testfile.foo"
echo "$TMPDIR"
realpath "$TMPDIR"

On Snellius, two jobs on node tcn386 used /scratch-local/scur1448.4401218 and /scratch-local/scur1448.4401219. realpath added /gpfs/scratch1/nodespecific/tcn386/ in front of the same job directory.

From episodes/files/ex1 on Snellius:

BASH

sbatch run.sh

On a small cluster:

BASH

iend=8 wdir=TMPDIR sbatch run.sh

iend=8 writes 10^6 and 10^7 values. wdir=TMPDIR is a keyword the script expands inside the job. It is not the login node’s $TMPDIR.

Glossary

metadata server: The part of a parallel file system that records names, owners, permissions, sizes, and where the bytes live. ls and creating a file talk to it.
data server: The part that holds the bytes. Reading a large file talks to it after one metadata lookup.
inode: One metadata record for one file. A quota can run out of inodes while bytes remain. Many small files spend inodes. One archive spends one.
home: /home/$USER. Small, backed up, the right place for scripts and small inputs. On Snellius the quota is 200 GiB.
scratch-shared: /scratch-shared on Snellius. Fast temporary space visible from the compute nodes. Unused data is removed after 14 days.
scratch-local: Job space on the shared scratch file system, reached as $TMPDIR. On Snellius the path looks like /scratch-local/$USER.$SLURM_JOB_ID. The table gives it 6 days.
scratch-node: NVMe on the node itself, on Snellius, requested with --constraint=scratch-node. Removed when the job ends. Not on every node. The course cluster does not have this constraint.
TMPDIR: The environment variable that points at this job’s scratch directory. It is a different directory on the login node and inside a job.
quota: The cap on bytes, and often on inodes. On Snellius, myquota prints it. On the course cluster, df -h and du -sh are the tools.
mode string: The ten characters ls -l prints for type and permissions. chmod g-r clears the group read bit.
project: Shared space that lasts for the life of the project. /projects on Snellius. The course cluster tab names its own project path.
archive: Slow, backed-up, long-term space on Snellius. The course cluster tab has no archive path.