User Tools

Site Tools


computing_drop

DROP cluster

  • Drop is:
    • drop.astron.nl
    • With 6 nodes at 8 cores each, DROP is designed for non-interactive queue-based parallel supercomputing.
  • Drop is not:
    • For anything that is interactive, uses less than 8 parallel cores per job, is not queue-able, or generally takes less than a few nights to run on your desktop, you had better use one of the stand-alone servers.
    • Drop has only scratch storage. Move important data to a different server. Anything on /data/ can be deleted without notice.
    • Kindly note that this cluster comes 'as is' with no support from IT, Joeri or Jason.
    • The login node 'drop' is also the file server. Do not run jobs there, in for- or background.
  • Come see Joeri for an account
  • File systems
    • /export/home/<user>
      • RAID6, 20GB/user quota, nothing ever auto-deleted
    • /data (crossmounted over all nodes) = RAID0, unlimited space
      • RAID0, 5TB, no quotas
      • heaviest users will have older data auto-deleted by semi-random script, so better clear it out yourself
  • Login/master/fileserver node is “drop”, nodes are called c0 to c5
  • Queueing implemented to balance loads. Only use queue, do not start processing on your own.
  • Cluster runs CentOS5 (a version of RHEL) under Rocks 5.1
computing_drop.txt · Last modified: 2009/05/06 17:37 by leeuwen