BLUEGENE - env vars such as SLURM_NNODES, SLURM_JOB_NUM_NODES, and SLURM_JOB_CPUS_PER_NODE now reference cnode counts instead of midplane counts.  SLURM_NODELIST still references midplane names.

    nodes already in another reservation.

  from the database

  This adds job preemption support to select/linear for preempt/qos and 
  optimizes job placement on nodes considering workload from other partitions.

svn merge https://eris.llnl.gov/svn/slurm/trunk https://eris.llnl.gov/svn/slurm/branches/slurm-1.4.srun.hang

I merged changes from the trunk onto slurm-1.4.srun.hang branch as I worked,
so the branch was up to date with ongoing trunk development.  The merge
command above finds the diffs between the trunk and branch--just the development
for the branch--and applies the changes to the trunk.

Not currently used

    suspended job is cancelled.

    Accounts/Users with no associations, Accounts with no users or subaccounts
    attached in a cluster, and Users with No UID on the system.
 -- Added new option for sacctmgr list assoc and list cluster WOLimits.  This 
    gives a smaller default format without the limit information.  This may 
    be the new default for list assocations and list clusters.
 -- Users are now required to have an association with there default account.
    Sacctmgr will now complain when you try to modify a users default account
    which they are not associated anywhere.

scheduling is performed.

This is a merge of all slurm modifications added to support a Sun Constellation system 
  with 4-D name space (3-D for coordinate followed by a sequence number for the node at
  that coordinate).

    running for a job allocated to that node.

    or too small, automatically try using the previously saved state (file 
    name with ".old" suffix).
This is the same logici added early for jobs.

    previously saved state.

+    ambiguity in the documentation over use of the terms "CPU" and "socket".