Date:	Tue, 11 May 1993 09:19:00 PDT
From:	minshall@wc.novell.com
To:	mobile-ip@parc.xerox.com
Cc:	ch@zurich.ibm.com, dupont@mdd.comm.mot.com (Pierre Dupont)
Message-Id: <9305111619.AB03368@wc.novell.com>
Subject: Re:  Beaconing

Hi, all.

I think that Colin Harrison and Pierre Dupont have basically the same view
i have of beaconing, but just to drive the point home once more...

Link level cell discovery/handoff should occur at the link level.  It would
be nice (as in Charlie's 802.11 proposal) if the link level would tell *us*
that a switch has occurred.

At the IP level, there are four (4) possiblities about which subnet an MH
might be in.  First, it may be on its "home subnet", in which case it
doesn't need to carry out any periodic action.  It needs to cancel any
current forwarding order with its home whatever-it-is-called (WIIC).

Second, it may be on a subnet which is willing to take responsibility for
routing packets destined for the home address to the active subnet ("host
routing", for example).  In this case, the MH will need to broadcast its
existence (on the subnet) periodically.  The MH will also need to cancel
any current forwarding order with its home WIIC.

Third, it may be on a subnet which is unwilling to take responsiblity for
routing packets destined for the home address, but requires a forwarding
address in order for packets to be delivered to the MH.  In this case, the
MH does not need to perform any periodic action, but *does* need to acquire
a forwarding address and instantiate a forwarding order with its home WIIC.

Fourth, like the second, is the case where the forwarding address is not on
the active subnet, and so the MH needs to periodically advertise its
existence in order to cause packets to find their way to the MH.

(Note that this is a 2-dimensional space - (need, don't need) to advertise
periodically; (need, don't need) a forwarding address.)

When an MH decides it (may) be on a new subnet (cell switch != subnet
switch, of course), then it needs to decide which case pertains.  My guess
is that the MH should multicast some request, and hope for a reply which
will allow it to decide what to do next.

In order to support link layers which do *not* notify about cell switches
(ethernet, for example), there should be some sort of beaconing at the
network layer which allows the MH to determine, at the minimum, whether it
has changed subnets.

HANDOFF

Link level handoff is a link level issue.

Network layer handoff, to me, is an appropriate issue for the network
layer.  There may be packets headed for the old subnet, and the old agent
near the old subnet should be informed of the move of the MH and asked to
forward packets to the new subnet.  Why?  Because while it is true that the
"network" can tolerate lost packets, everything works better if fewer
rather than more packets are lost.  If we can do something fairly simple to
reduce the number of lost packets, that is a win.  (Note that we can
probably only do this in the case where the link layer is willing to tell
the network layer when a cell switch has occurred.)

(What i really don't know how to make "regular" is how to integrate in the
idea that you might, especially during handoff, be connected to more than
one base station at a time.)

Greg Minshall