There are three classes of disk storage on the cluster: home directory, scratch, and project-associated mass storage.

Home directory storage is located at /home. Storage use is limited to 40 GB per user by default but can be increased on request. This file system is globally available and is intended to house source code, binaries, and other relatively small files. Data in the home directories is backed up nightly (see Data Backup). Users may check their home directory utilization using the "df" command: df -h ~/.

There are two types of scratch space: local and global. Each worker node has a locally attached disk that is available for temporary storage during the execution of a job. The location of this directory is /tmp/. Available space is on the order of 100GB. Access to this space is only possible during the execution of the job on that compute node. Local scratch storage is not backed up.

Global scratch space is available at /lustre/scratch. This is a single storage space that is visible on all cluster nodes. To use this space you should create a subdirectory here using your username, and store files under this directory. Files which have not been modified in the past month may be automatically deleted from this storage location.

For computing projects that have large storage requirements we offer larger storage allocation on the Lustre filesystem (under the /lustre/projects/ directory), the Gluster filesystem (under /omega/), and NFS shares (under the /data/ directory). You may apply for use of this storage on the HPC web manager page. Backup of this data is available with special arrangements.

The /data directory on cluster machines is a special mount point that is used by the autofs service. Directories that are located under this mount point may not be visible until an attempt is made to access them by either reading a file under the directory or using cd to enter the directory structure. This is because these directories are mounted on an as-needed basis.

Home directory files of size 10GB or less are backed up to a remote server located across campus. The schedule for backups is roughly daily as transfer limitations allow. In addition, the backup data is snapshotted daily to allow users to go back to previous file versions and to undelete files as needed. The amount of historical data that is retained is determined by the available space on the backup server.

To access the backups, users should change to the backup filesystem (/data/backup/) and choose the snapshot date that they wish to access. Once the files are located, the user should copy the files to their active storage area on the cluster. All backup data is read-only. Here is an example of retrieving a deleted file from the home area:

[jdoe@newton ~]$ cd /data/backup

[jdoe@newton backup]$ ls
2011-12-14  2012-01-08  2012-02-02  2012-02-27  2012-03-24  2012-04-18
2011-12-15  2012-01-09  2012-02-03  2012-02-28  2012-03-25  2012-04-19
2011-12-16  2012-01-10  2012-02-04  2012-02-29  2012-03-26
2011-12-17  2012-01-11  2012-02-05  2012-03-01  2012-03-27
2011-12-18  2012-01-12  2012-02-06  2012-03-02  2012-03-28

[jdoe@newton backup]$ cd 2012-04-19

[jdoe@newton 2012-04-19]$ cd $USER

[jdoe@newton jdoe]$ ls
lsb.acct                     test.sh.e0
lsb.hosts                    test.sh.e1147
lsb.params                   test.sh.o
lsb.queues                   test.sh.o0
lsb.resources                test.sh.o1147
lsb.users                    tmp
man.strace                   torque-2.1.9
man.strace.1                 torque-2.1.9.tar.gz
man.strace.root              users.pl
man.strace.root.1            users.txt

[jdoe@newton jdoe]$ cp lsb.queues ~/