Filesystems

From HPCwiki
Jump to navigation Jump to search

Fast Scratch

On the Lustre PFS scratch space is organised per partner. Users can only create directory and files in the folders of the organisation they belong to. The Fast Scratch is meant for temporary files and folders. Files and folders should be removed once the jobs are finished. Files and folder older than one month will automatically be removed. Since the Fast Scratch is in an integrated part of the compute infrastructure, no additional cost is incurred based on use in either throughput or volume stored.

 /lustre/scratch/[partner]/[unit]

e.g.

 /lustre/scratch/WUR/ABGC/

Fast Protected

Data that needs to remain on the Lustre PFS *and* needs to be backed up as well (i.e. requires redundancy in case the PFS experiences a fatal failure) can be placed in the Fast Protected area.

 /lustre/backup/[name partner]

e.g.

 /lustre/backup/WUR/ABGC/

Note that this map will not be backed up yet. It is planned that daily syncing will commence 1-1-2014.

Fast Unprotected

The Fast Unprotected area is meant for data that needs to remain on the PFS, but that otherwise is not so valuable that it requires regular backup. This could for instance include results computationally derived from primary data sources that are backed up, and for which it is computationally feasible to reconstitute it in case of a FS failure. The advantage of using the Fast Unprotected over the Fast Protected is cost: the absence of the backup function means generally halving of cost.

 /lustre/nobackup/[partner]/

e.g.

 /lustre/nobackup/WUR/ABGC/

fast shared

compute nodes scratch