and Moe Jette. There are still a couple of end-cases that need to be fixed. We also
  need to push the core identification out to the slurmd for task binding. sched/gang
  needs revision to support this too.

  enabled. Only effects sched/backfill, but make sure the thread is 
  finished before purging job data structures

structure in preparation for removal from slurmctld job_record

    a job's completion before allocating already released resources to pending
    jobs. This can be used to reduce fragmentation of resources.

  (batch launch within existing allocation).

  documentation, read config, show config, and run PrologSlurmctld when a job
  is allocated resources (startup synchronization still needed).

    https://computing.llnl.gov/linux/slurm/power_save.html or 
    "man slurm.conf" (SuspendProgram and related parameters) for more 
    information.
This is the final installment of the work: update some documenation, 
    increase default ResumeRate, and reduce frequency of retrying 
    batch launch state check.

that it doesn't get set DOWN right away for not responding since
the last time it was powered up.

Don't allocate nodes to a job step until it is responding (as before) AND
  the node is no longer in power save mode.

    "man slurm.conf" (SuspendProgram and related parameters) for more 
    information.
NOTE: the step create logic needs to be modify to return EAGAIN or the 
    like until the node's state NOT_RESPONDING flag gets cleared.

    longer needed.