GEOCEAN — Data & Storage

How to use the different filesystems available to the GEOCEAN group on the SCI-UC cluster. See also the general SCI-UC storage docs for cluster-wide policy (this page only covers GEOCEAN-specific usage).

Overview

Path Backend Quota Use for
/nfs/home/geocean/$USER NFS 120GB / user dotfiles, scripts, small configs, source code, conda envs
/lustre/geocean/WORK/$USER Lustre 10TB, 1M files active job I/O, intermediate model output
/lustre/geocean/DATA Lustre 10TB, 1M files (shared) shared, long-lived group datasets
/nfs/software/geocean/ NFS — group-built software, modulefiles, shared conda envs
/scratch (per node, job-local) local NVMe/SSD — fast scratch for a single running job

/nfs/home/geocean/$USER — home directory

Your home directory. Good for:

  • dotfiles, shell config
  • small scripts and source code
  • conda/venv environments

Do not store large datasets or run I/O-heavy jobs here — it's NFS, not built for that, and it's the most common way to slow down everyone's home directory at once.

/lustre/geocean/WORK/$USER — your working space

Use this for anything a running job reads or writes: intermediate files, model output, checkpoints. This is high-performance parallel storage meant for active work, not long-term keeping.

Check your usage:

lfs quota -hu $USER /lustre/geocean/WORK

/lustre/geocean/DATA — shared group data

For datasets shared across the group long-term: forcing data, reanalyses, finished model outputs others will reuse. This is not scratch space — coordinate with whoever else uses a given dataset before deleting or restructuring things.

/scratch — per-job local storage

Each compute node has local NVMe/SSD mounted at /scratch, available only while your job is running.

  • Exists only on that node, for that job
  • Not backed up
  • May be deleted without notice once the job ends

Copy any results you need out to /lustre/geocean/WORK/$USER before your job finishes, or you will lose them.

Backups

Daily snapshots, 7-day retention, cover:

  • /lustre/geocean/WORK and /lustre/geocean/DATA
  • /nfs/home/geocean and /nfs/software/geocean/

To request a recovery from a snapshot, contact soporte.sci@unican.es within the 10-day window.

Quick rules of thumb

  • Small stuff, code, configs → /nfs/home/geocean/$USER
  • Job I/O while it's running → /lustre/geocean/WORK/$USER
  • Long-lived shared datasets → /lustre/geocean/DATA
  • Fastest possible scratch, single job only → /scratch
  • Never treat /scratch or /nfs/home as a place to keep something permanently

Open items

  • [ ] Document the group's convention (if any) for organizing /lustre/geocean/DATA — subfolder naming, ownership.
  • [ ] Add guidance on moving data between /lustre/geocean/WORK and /lustre/geocean/DATA once a run is finished.