Filesystems¶
Hydra has public partitions and a number of project partitions, each with its own quotas and retention rules. We adjust sizes and quotas as needed. disk-usage -d all+ -quotas on a login node prints the current values (Check your disk usage and quotas).
Public partitions¶
| Partition | System | Size | Space quota, soft / hard | File quota, soft / hard |
|---|---|---|---|---|
/home |
NetApp | 22 TB | 350 GB / 384 GB | 9 M / 10 M |
/data/public |
NetApp | 330 TB | 4.3 TB / 4.5 TB | 9.5 M / 10 M |
/scratch/public |
GPFS | 800 TB | 14 TB / 15 TB | 37 M / 39 M |
| Partition | Snapshots | Scrubbed | Use for |
|---|---|---|---|
/home |
4 weeks | no | configuration files, scripts, job files, source code |
/data/public |
2 weeks | no | final results, small important files; space from deleted files is not freed until the snapshots age out |
/scratch/public |
none | files older than 180 days, weekly | input data and working files for jobs; the fastest partition, over InfiniBand |
/store/public is being phased out
/store/public is no longer available. Use /data/public. A cloud-based cold-storage service is being evaluated as the replacement. Project /store partitions are not affected.
A soft limit can be exceeded for a grace period. At the hard limit writes fail. The file quota counts inodes, so many small files reach it before the space quota does. /scratch/dbs (10 TB) holds the shared bioinformatics databases. See Local databases.
/scratch/public is divided by unit or discipline into biology, genomics, humanities, nasm, odi and sao. Your directory is under one of them, /scratch/public/genomics/USERNAME for example. The shorter form /scratch/genomics is a link to /scratch/public/genomics, and either works. On /data, /data/genomics is the same directory as /data/public/genomics.
Project partitions¶
Groups with their own funding have dedicated partitions under /scratch and /store, with quotas set per project. quota+ shows the ones you have access to. Dedicated space is bought with project funds when the disk farm is expanded. Email [email protected].
Storage systems¶
| System | Partitions | Access | Properties |
|---|---|---|---|
| NetApp | /home, /data |
every node | snapshots; disaster-recovery copy to AWS Glacier (see Backups) |
| GPFS over InfiniBand | /scratch |
every node | fastest; no snapshots; scrubbed |
| NAS | /store (project partitions) |
login, head, interactive and RStudio server nodes only | snapshots on most partitions, for 2 to 8 weeks; fault tolerant but not highly available (see Use /store and the I/O queue) |
| local SSD | $SSD_DIR |
the node the job runs on, while it runs | see Use a node's local SSD |
Scrubbing¶
Files on /scratch/public older than 180 days are deleted every week
/scratch has no snapshots and no backup. Move anything you need to keep to /data or off Hydra before it is 180 days old.
A scrubber runs weekly and removes files on /scratch/public older than 180 days, and old empty directories. It holds the removed files in a staging area for about ten days before deleting them. During that time they still count against your quota, and we can restore them on request. See Find scrubbed files and request a restore. A restored file's change time (ctime) is reset, so it is safe for another 180 days.
None of the storage systems on Hydra are meant for archival storage. Delete what you no longer need. Compress or archive sets of small files. tar -czf archive.tgz dir/ packs a directory into one file, and tar -xf archive.tgz unpacks it.
Last updated 2026-10-03