Beginning Tuesday, January 11 at 09:00am, the Hyak job scheduler will undergo a number of changes to improve cluster reliability, performance, and utilization. Below you will find a summary of changes in behavior that will affect users.

MAJOR CHANGES

  1. Node local scratch disks (n####:/tmp, n####:/scr, n####:/var/tmp) will be cleaned up after the completion of each job. Groups that would like to keep data on the disks in their nodes should save any data into a directory with their group name. If your UNIX group is hyak-mygroup, your directory should be named mygroup.
  2. Jobs must use all the processors on a node. For many users, this will require no changes. If your jobs use fewer than 8 processors, you will need to create job scripts that start multiple jobs. Sample scripts are available on the Hyak User Wiki: (Hyak_Serial_Job_Scripts).
  3. Users will now be able to submit to a special queue, bf, and those jobs will run on the 56 nodes owned by the eScience Institute. Users can specify an alternate queue using the -q msub option, msub -q bf myjob.sh. These jobs can be canceled and requeued by eScience job submissions, so short jobs or jobs that use checkpointing are best suited to this new queue. Interactive jobs cannot be run in the bf queue.

OTHER CHANGES

  1. When a job is submitted, there may be a delay of up to thirty seconds before the job is scheduled. This delay will be visible when running showq or starting an interactive session.
  2. Users who submit hundreds of jobs at once should use qsub in place of msub.
  3. A resource called file will now be available for each node indicating how much space is available on the local scratch disk. msub -l file=50GB myjob.sh would only select nodes that have at least 50GB free on their local scratch disk.
  4. Users with access to multiple groups' nodes will now have to specify their group, rather than the partition name when they are submitting jobs to their non-default group/partition. Most users are only in one group and will not have to change their behavior. If your default UNIX group is hyak-mygroup and your secondary group is hyak-othergroup you'll have to run qsub -W group_list=hyak-othergroup myjob.sh if you'd like to use the nodes owned by othergroup. You must also use qsub instead of msub.