Storage Options#
For comprehensive details about the storage systems available on Aire, please refer to the Storage and Filesystem section.
Using Flash on Lustre (NVMe‑based)#
Flash on Lustre provides high-performance, NVMe-based temporary storage, ideal for I/O‑intensive workloads and jobs requiring rapid access to large datasets across multiple nodes. The path to this storage is set automatically in your job as the environment variable $TMP_SHARED, and typically points to /mnt/flash/tmp/job.<JOB-ID>.
To utilise Flash storage in your job, reference $TMP_SHARED in your submission script. For example:
#!/bin/bash
#SBATCH --job-name=example
#SBATCH --time=01:00:00
Flash storage path is automatically set as $TMP_SHARED
echo "Flash storage path: $TMP_SHARED"
Copy input data to Flash storage
cp -r /path/to/input/data $TMP_SHARED/
./example.bin --data $TMP_SHARED/data
Copy results back to permanent storage
cp -r $TMP_SHARED/results /path/to/permanent/storage/
Note
Data stored in Flash is automatically purged when the job finishes. Always copy important results back to permanent storage before your job ends.
Using Scratch on compute nodes#
Scratch storage on compute nodes is accessed via the environment variables $TMP_LOCAL or $TMPDIR, which typically point to /tmp/job.<JOB-ID>. This storage is ideal for single-node jobs that require fast, localised access to temporary data.
To utilise it, reference $TMP_LOCAL or $TMPDIR in your submission script. For example:
#!/bin/bash
#SBATCH --job-name=example
#SBATCH --time=01:00:00
# Local scratch storage path is set as $TMP_LOCAL or $TMPDIR
echo "Local scratch path: $TMP_LOCAL"
# Copy input data to local scratch
cp -r /path/to/input/data $TMP_LOCAL/
./example.bin --data $TMP_LOCAL/data
# Copy results back to permanent storage
cp -r $TMP_LOCAL/results /path/to/permanent/storage/
Note
Data stored in $TMP_LOCAL or $TMPDIR is automatically deleted when the job ends. Ensure you copy any important results back to permanent storage before your job completes.