Filesystems and Data Handling
Available Filesystems
Listed below are the file systems available on UBT's HPC systems that may relevant to you.
Please review the following table carefully! This applies in particular to the data retention periods, snapshots and backups. Once the data retention period has expired or the data has been deleted from a storage medium without a backup or snapshot, there is no chance to bring your data back.
| mountpoint | cluster | purpose | FS Type | backup/snapshots | data lifetime | quota |
|---|---|---|---|---|---|---|
| /home/ |
ALL | configuration data, scripts, small input data | NFS | yes/yes | Cluster EOL / Account Lifetime | 10GB |
| /groups/ |
Festus | large input and output data for contributing groups | NFS | yes/yes | Cluster EOL | contribution dependend |
| /workdir/$USER ($WORKDIR) | Festus | workspace Festus | NFS | no/no | 60days | 3TB |
| /scratch/$USER ($SCRATCH) | Festus | High performance parallel IO | BeeGFS | no/no | 10days | No |
| /scratch/$USER | Lipwig | High performance parallel IO | PanFS | no/no | Watermark | no |
| /tmp | ALL | local temporary dir(job runtime) | XFS | no/no | job | no |
To make use of High Perfomance FS and securing your data it's recommended to take a look at our data staging example.
Transferring Data
This page shows how to move files between your own computer, server and the cluster. It covers scp and rsync. We'll use the Festus Cluster as an example, but this will work with other systems as well.
Prerequisites
Before using scp or rsync, make sure you can log in to the login server:
ssh <user>@festus.hpc.uni-bayreuth.de
If this works, the copy tools will work too. You will be prompted for your password every time you start a transfer, unless you have set up SSH keys.
scp
scp copies files over SSH, either from your computer to the cluster or the other way round.
General form:
scp [<options>] <source> [<source> ...] <destination>
You can give several sources. Source and destination are each either
- a local path, or
- a path on another machine in the form
<user>@<host>:[<path>]. - Without
<path>, the remote home directory ($HOME) is used. - A relative path (no leading
/) is resolved relative to$HOME, sodatabecomes$HOME/data.
Common options:
| Option | Effect |
|---|---|
-r |
Copy directories recursively (required for directories) |
-p |
Preserve modification times and permissions |
From your computer to the cluster
scp -r sim09.in ~/archive/run_07 /mnt/external/rawdata <user>@festus.hpc.uni-bayreuth.de:incoming
The three directories sim09.in, ~/archive/run_07 and /mnt/external/rawdata all end up in $HOME/incoming on the cluster. Because incoming is a relative path, $HOME is prepended automatically. Since these are directories, -r is required.
scp -r model_v2 <user>@festus.hpc.uni-bayreuth.de:/scratch/<user>
The directory model_v2 is copied into your scratch directory. Adjust the absolute path to match your own directory.
From the cluster to your computer
scp -r <user>@festus.hpc.uni-bayreuth.de:/scratch/<user>/results/2024 /mnt/backup/
The directory /scratch/<user>/results/2024 is copied to /mnt/backup/ on your local machine.
Wildcards in remote paths
Asterisks (*) and question marks (?) in remote paths must be escaped with a backslash, otherwise your local shell tries to expand them:
scp <user>@festus.hpc.uni-bayreuth.de:/scratch/<user>/results/2024/\*.csv .
This fetches all .csv files from /scratch/<user>/results/2024 into your current local directory.
rsync
rsync serves the same purpose as scp but can do more: it can transfers only what has changed and can resume interrupted transfers. For large amounts of data or repeated syncing it is usually the better choice.
rsync [<options>] <source> [<source> ...] <destination>
Source and destination follow the same rules as for scp (a local path or <user>@<host>:[<path>]; relative remote paths refer to $HOME).
Common options:
| Option | Effect |
|---|---|
-a |
Archive mode: preserves permissions, timestamps, symlinks, etc. (already includes -p) |
-v |
Verbose output |
-P |
Show progress and keep partially transferred files if interrupted |
--append-verify |
Resume partially transferred files instead of sending them again |
-n / --dry-run |
Only show what would happen, without copying anything |
-z |
Compress data during transfer |
--compress-choice=<alg> |
Choose the compression algorithm |
--compress-level=<n> |
Choose the compression level |
Note on compression: Between systems inside the cluster network,
-zcan actually slow a transfer down, because the connection is already fast and compressing only costs CPU time. It is mainly useful over slow links, such as from your home network.
Local directory to the cluster (mind the trailing slash!)
rsync -avz samples/series_a <user>@festus.hpc.uni-bayreuth.de:samples
The directory series_a itself is created: a local file samples/series_a/sample1.dat ends up as $HOME/samples/series_a/sample1.dat.
rsync -avz samples/series_a/ <user>@festus.hpc.uni-bayreuth.de:samples
With a trailing slash on the source, only the contents are copied: the file samples/series_a/sample1.dat now ends up as $HOME/samples/sample1.dat.
Rule of thumb: No trailing slash = the directory itself is copied. Trailing slash = only what is inside it.
From the cluster to your computer
rsync -avz <user>@festus.hpc.uni-bayreuth.de:/scratch/<group>/<user>/output ~/analysis
The directory output is copied to ~/analysis and will be located at ~/analysis/output.
rsync -avz <user>@festus.hpc.uni-bayreuth.de:/scratch/<group>/<user>/output/\*.csv ~/analysis
Only the .csv files from output are copied to ~/analysis.
Resuming an interrupted transfer
If the connection drops, simply run the same command again, preferably with -P or --append-verify:
rsync -avP --append-verify large_data.tar <user>@festus.hpc.uni-bayreuth.de:incoming/