Suggested Use

How you organise your files between home and scratch is up to you. However, one suggested method is to create the same directories in each area and then to keep source code, input files, and other important files you want backed up in the directories in home. As an example, you could name your directories in home and scratch based on the experiment and date you are running on, for example:

mkdir -p ~/heat_tfer/material05/run01_130625/
mkdir -p ~/scratch/heat_tfer/material05/run01_130625/

To actually run your jobs you can copy the input files, jobscripts, compiled executables etc from the home area to the equivalent scratch directory (large code compilation could be done directly in scratch). Run your job from scratch (which usually means running the qsub command from within the scratch area) and then copy important results back to home once the job has finished.

Remember, scratch is a high-performance file system so can reduce the overall time your job takes if your code reads and/or writes large (or many) files. We’ve seen this reduce job times by hours in some cases. It is also much larger than your home area (which is subject to quotas) so is generally safer to run in scratch in case your job creates any large temporary files – if your home area goes over quota the job will fail.

But please keep your scratch area tidy – delete unwanted job files once the job has finished (many apps create temporary files which may not be automatically deleted by the app). Leaving thousands of files in your scratch area will slow down your access to that area. By regularly moving important results to the home area (where they are backed up) and deleting unwanted files you’ll get the best performance from the filesystem and find managing your files a lot easier.