BUGS
----

- Job.submit (PrintStream) has a race
- Cluster.exit () is untested
- NodeImpl: force to unload class after execution


FEATURES
--------

- use EventMailbox for output

- WildcardName (Entry for node names)
    - may have to push this into NodeImpl during node discovery/alloc
- RemoteThreadGroup
- file access
- configuration
    - shell scripts
    - test/SubmitManual.java
    - com/nukernel/jdcf/tools/Submit.java
- implement Job's suspend, resume, and kill methods
- job object's lease needs to be distributed to all machines.  right now
  only one has it, and that machine failing would cause its lease on the
  LeaseRenewalSet to expire which would cause the lease on the object in
  the JavaSpace to expire.  we probably just need to bite the bullet and
  make every node participate in every job object update so that we don't
  have to bother with any updating nonsense.

LONG TERM
---------

- dynamically add nodes

THINK
-----

- should NodeImpl lookup JavaSpace again since it already is in the Job
  object?  it does now and is transient in Job.

DONE
----

- Janitor (a lease renewal service that renews perpetually)
      [ not needed.  the service jini provides `leases' it's renewal,
	  but it *always* honours the requested lease length, even if it's
	  forever (not recommended, though; use a day or two instead). ]
- use LeaseRenewalService

