Overview of available file systems
File system + mount point Environment Variable for Access | Purpose | Type and Size | Backup and snapshots | Intended lifetime, cleanup strategy | Quota |
|---|---|---|---|---|---|
/home $HOME | Store source files, input data, small files Globally accessible from login and compute nodes | GPFS, 100TB | backup to tape and snapshots | Account lifetime | 100GB per user |
/work $WORK | Result files and data sets Globally accessible from login and compute nodes | GPFS, 100TB | Backup upon request, at cost price. Snapshots | 3 years | 50GB* per group |
/scratch $SCRATCH | Temporary huge result files Accessible from the frontents and the compute nodes within one cluster. | Deneb: GPFS 350TB Fidis: GPFS 375TB | No backup, no snapshots | 2 weeks Automatic deletion of files and empty directories older than 2 weeks may happen without notice | Quotas may be in place to prevent runaway tasks filling the filesystem. A typical limit would be 1/3 or 1/2 of the total space. |
/tmp/${SLURM_JOB_ID} $TMPDIR | Temporary, local file space for jobs on compute nodes. Not available on login nodes. | ext4 Varies between 64 and 512 GB | No backup, no snapshots | job execution content is deleted after job end | no |
* Space on work is charged for and, as such, the quota of the group depends on the amount of space purchased. There is no backup by default but a laboratory may request such a service. The price for backup will be the cost price to SCITAS.
The price for work is 300 CHF per TB for 3 years.
Each group is allocated, for free, 50GB.
Security of user data
The contents of the home file-system are backed-up on a daily basis with a six month retention period. The backed up data are held at a separate physical location to the original data.
The contents of the work file-system are not backed up by default
The scratch file-systems are not backed-up under any circumstances.
The scratch file-systems are only for short-lived files and, in the case of insufficient free space, files older than two weeks may be deleted without notice in order to ensure the usability of the cluster.
Scratch automatic cleanup
When a scratch filesystem reaches a certain level of use (normally 90%) an automatic cleanup procedure is activated. Starting with the oldest files present deletion takes place until the occupancy has been reduced to 70%. Only files less than 2 weeks old are sure not to be deleted by the cleanup mechanism.
Files belonging to a former user
When a user no longer has a valid account on the clusters any files belonging to him on home are removed from the servers. They will remain on tape for 6 months after the user has left EPFL. The head of the laboratory is responsible for ensuring that these data are correctly managed. He or she can ask for a retrieval from tape.
The work file-system is divided by laboratory and, as such, it is the responsibility of the head of the laboratory to ensure that the data are correctly managed.
Once a user is no longer accredited, files belonging to them in scratch can be deleted without notice.
How to recover snapshots
A snapshot is the state of a system at a particular point in time. On our clusters, the home and work filesystems are snapshotted daily and snapshots are kept for one week. This is particularly useful in case a user removes a file by mistake.
Daily snapshots of the home and work filesystems can be found in /home/.snapshots, /work/.snapshots, respectively.