Return-Path: <ghudson@MIT.EDU>
Received: from pacific-carrier-annex.mit.edu by po11.mit.edu (8.9.2/4.7) id NAA26623; Fri, 28 Sep 2001 13:36:46 -0400 (EDT)
Received: from egyptian-gods.MIT.EDU (EGYPTIAN-GODS.MIT.EDU [18.101.0.162])
	by pacific-carrier-annex.mit.edu (8.9.2/8.9.2) with ESMTP id NAA01345;
	Fri, 28 Sep 2001 13:33:24 -0400 (EDT)
Received: (from ghudson@localhost) by egyptian-gods.MIT.EDU (8.9.3)
	id NAA14071; Fri, 28 Sep 2001 13:33:23 -0400
Message-Id: <200109281733.NAA14071@egyptian-gods.MIT.EDU>
To: John Hawkinson <jhawk@MIT.EDU>
cc: Garry Zacheiss <zacheiss@MIT.EDU>, sipb@MIT.EDU
Subject: Re: On zephyr ghosts
In-Reply-To: Your message of "Fri, 28 Sep 2001 13:23:35 EDT."
             <20010928132335.I23491@multics.mit.edu> 
Date: Fri, 28 Sep 2001 13:33:23 -0400
From: Greg Hudson <ghudson@MIT.EDU>

> This explanation doesn't really seem to hold water for me. Do you
> really believe it? That the zephyr server frequently enough drops
> udp packets indicating that the "connection" is done? I thought
> there was some sort of liveness probe anyhow?

Dropped packets shouldn't be a common cause of zephyr ghosts, since
zhm will retransmit.  The most easily explained cause of zephyr ghosts
is a zwgc process which dies uncleanly or a machine which reboots
uncleanly.

If your client has subscriptions in addition to a location (the common
case), your location will go away when your subscriptions go away.
That takes about 17 minutes, though, and the server has to have tried
to send you a notice to start that timer.  So a user who doesn't
receive notices very often could have a zephyr ghost lying around for
a long time.
