Home >

Linux Cluster (rcluster)

Disk Space Usage on the rcluster

Home directory:

The directory that you access when you login to rcluster is your home directory /home/groupname/username, which has a default quota of 1GB. This is the same home directory that you access when you login to other RCC resources, such as the IBM pcluster and the SGI altix.  To see what your disk quota is (in KB) and how much disk space you currently have allocated use the command:

quota   -v

If you need more disk space for your home directory, please refer to the RCC policy page.

---

Scratch space:

Users can submit their jobs from the scratch area, located at /scratch/username. There is currently no user quota for the scratch area. Files that are no longer needed for active jobs should be transferred elsewhere (to your home directory or to your local machine). An automatic purging policy for files in /scratch may be established in the future if this area fills up with old files.

---

Local temporary space:

Each compute node has approximately 45GB of temporary disk space called /lscratch that cannot be directly accessed by the login node rcluster.rcc.uga.edu. Since /lscratch is local to each node, I/O to it from the node is very fast. Temporary files generated by jobs can be written to it (use the subdirectory /lscratch/username ). If your job is I/O intensive, you may consider having your job submission script copy files to /lscratch/username before running the program. However, files in /lscratch are automatically purged. Therefore, it is advisable that your job submission script transfer relevant files out of /lscratch/username before your job ends. The current purging policy implemented is that files that are older than 60 days and not accessed or modified in the past 30 days are deleted. This grace period may be reduced in the future if /lscratch gets too full and thereby not usable by new jobs. Note that because /lscratch is local to each node (and not visible to other nodes), you must use this space with care if running parallel jobs with multiple nodes.

Example

Here is a job submission script that copies the inputfiles (input1.txt and input2.txt) and the serial executable (a.out) from the user's home directory (e.g /home/groupname/username/subdirectory) to /lscratch/username and then runs the executable from that directory. The output files (output1.txt and output2.txt) are written to the local scratch area as well, and then they are moved back to the user's home directory. At the end of the script, files that are no longer needed (such as the executable a.out and the input files) are deleted: