Filestore

The new cluster has a separate central filesystem from the old. To make life easier, user data has already been copied over on to the new system.

User filestore (home and storage) was initially synchronised at the begining of October. If you've used the old system since that point, the contents of your directories on the new system will likely differ. Copying across new files form the old system is a simple two-stage process: file synchronisation followed by group conversion.

• File synchronisation

  The best tool to use to synchronise files from the old system with the new is rsync. For your home directory, type the following on the old HPC's frontend, corona:

  rsync -av /home/hpc/xy/username/ elysium.hec.lancs.ac.uk:/home/hpc/xy/username

  And for your storage filestore:

  rsync -av /storage/hpc/xy/username/ elysium.hec.lancs.ac.uk:/storage/hpc/xy/username

  Where xy is the bin number of your account, and username is your username.

  Please Note: The trailing slash on the first directory argument is important.

  Synchronisation can be a slow process; if you wish to transfer only a small set of files across to the new cluster, you may wish to use rsync on individual directories within your account, or a file transfer tool such as scp.

• Group conversion

  Old style usergroups such as st (staff) have been merged together into a new group called local. All local users will have this new group as their primary one. rsync and many other file transfer methods will retain a file's old user group when moving files across. This can be fixed using the gconv tool. To convert all files within your home directory, type the following on the new cluster's login node elysium:

  gconv /home/hpc/xy/username

  and for your storage area:

  gconv /storage/hpc/xy/username

  gconv accepts any single file or directory as an argument, so it can be used if you want synchronise just a subset of your filestore too.

Changes

There are a number of changes in the way the new system works. Here's a list of things to bear in mind:

• Job scheduler: The new system uses LSF rather than SGE for job scheduling. Job submission and monitoring commands, as well as job submission script syntax are all different. The relevant pages have been updated with information on these new commands and formats
• Binaries: The new system is generally binary compatible with the old, but we strongly recommend recompiling your code to make the most of newer CPU features.

    
  

  ©Lancaster University   Computer User Agreement   Privacy Statement