Go to the U of M home page
School of Physics & Astronomy
Physics & Astronomy Wiki

computing:department:unix:file_storage

Data storage on Unix

Home directories

Your home directory on the Unix cluster has a usage quota, to avoid too much space being taken by individual users. The standard quota is normally somewhere around 10GB. You can see your current usage and quota by visiting MyPhys.

If you run into your quota, and after reviewing your file usage find that you still need more space, you can reply to the warning email to request a change from us. Please understand, however, that the amount of space available is limited. Your home directory is not intended for large research data sets, for which separate project-specific storage should be used.

Home directories should not be used for data-intensive computations, such as input or output for condor batch jobs.

The home directories are backed up nightly, and in addition filesystem “snapshots” are created every few hours (currently at 09:00, 12:00, 15:00, 18:00).

Shared project areas

If you need a shared area for a group or project where multiple people can develop or store code, we can create such an area and back it up for you. These are made available under the /local hierarchy. These are intended for relatively small-scale usage, such as shared program areas - large data sets should be stored in the research data areas.

Research data storage

Other file systems are provided for research or project-specific data, under the /data hierarchy. This storage space is purchased by the research group. It can take the form of simple single drives in linux workstations (though this is discouraged), part of a shared research storage pool, or dedicated systems for large-scale storage needs.

These file systems are usually named either after the research group group (for a fileserver volume), or with the name of the workstation which hosts it, and contain further directories organized by user or by project. These areas should be used for large data sets and storage for local processes. Note that these areas are provided by the automounter - they are not activated until they are first accessed, so they won't necessarily appear in the output of commands like df.

Backups

  • Unix home directories are backed up daily.
    • Older files may need to be recovered from snapshots or from tape. Complete tape backups are currently made monthly and retained for 6 months.
  • Research data areas are not backed up
    • Data areas on our ZFS storage have nightly snapshots which are kept for 2 days - these can help you recover from accidentally deleted files.
    • Linux RAID storage does not have snapshots.
    • If you have critical research data which requires backup, this can be arranged for moderate data sizes (eg < 10TB). Please talk to us about the options.
    • Local data drives in workstations are never backed up.

Local "scratch" storage

“Scratch” space is space that is not backed up, and generally only used for temporary storage.

Your scratch directory on the local system is always named /scratch/local/username, and can be accessed using the environment variable $SCRATCHDIR. Since it's directly connected to the computer you're using, access to data on it is generally faster than to network storage (such as your home directory). This can make it a good choice for processing bulky data. However, don't place any files you may want to keep long-term here - files which have not been accessed for 30 days or more may be purged from this area, or when a workstation is updated.

When running jobs under Condor, the environment variable CONDOR_SCRATCH_DIR gives the name of a directory where the job may place temporary data files.

Temporary directories

Please don't use /tmp for creating temporary files, if you can help it. This area is very limited in size, and filling it may cause problems for the system.

We define an environment variable TMPDIR, which points to a suitable (and larger) area for temporary files. Also consider whether SCRATCHDIR (described above) might be more suitable…

Remote access to data

The security model of Unix NFS is too weak to permit exporting it to any systems which are not part of the Physics cluster. Some alternative methods of transferring data include…

globus

sshfs

You can use fuse (the userspace filesystem driver) to mount any remote filesystem which you have ssh access to. Although this is not a high-performance solution, you can use this to access files from non-Physics linux systems. For example:

mkdir ~/mnt  #create a place to put it, can be called anything you want.
sshfs username@remotesystem:/path/to/mount ~/mnt

later, to unmount it:

fusermount -u ~/mnt
computing/department/unix/file_storage.txt · Last modified: 2016/02/02 13:52 by allan