2001-03-07 now works; below describes problem...

----------------------------------------------------------------------
From keen@cs.ucdavis.edu  Mon Dec 18 12:57:58 2000
Date: Mon, 18 Dec 2000 12:56:35 -0800 (PST)
From: Aaron Keen <keen@cs.ucdavis.edu>
To: olsson@cs.ucdavis.edu
Subject: Re: ~/vsuite/vm/invoke

  >I'm not clear.  The client doesn't return from the call because the
  >scheduler doesn't get the go_ahead.  So, the program should hang. I'm
  >confused about the VM "dies while returning from the call" -- I don't
  >see that it is.  Isn't it an implementation problem, not a user
  >problem?  (Is that what you meant above?)

	Yes, I believe that it is an implementation problem.  Here's
what I think is happening.

	1)  There is an outstanding RMI call to the client method.
	2)  The timer goes off signaling a shutdown.
	3)  JR.exit is called so it begins to notify each of the
		VM's to kill itself.
	4)  The VM on which the client object resides kills itself
		before the calling VM.
	5)  The calling VM, which is waiting for the reply from the
		RMI call, receives a truncated (perhaps completely
		empty) response.  (Probably a broken pipe or some
		such system-level error signaling a broken connection).
	 5a) This broken connection triggers an exception before that
		VM terminates itself.

	This is mostly speculation but it's my best guess at present.

