Re: Host Addressing Conventions [addr-conv 2.8]

[[e-mail]]

<<On Tue, 15 Jun 93 14:04:03 EDT, [e-mail] (Paul Francis--formerly Tsuchiya) said:

[Sorry for all the quoting, but it's needed for context.]

>>  
>>  Here's a diagram that I drew on the dinner table last night:
>>  
>>                              area1 | area2
>>        +---------+                 |
>>      +-+ router1 +-+               |
>>      | +----+----+ |               |
>>  sn1 |      |      | sn2           |
>>      |  sn3 |      |               |
>>      |      |      | +---------+   |   /other
>>    +-+--------+    +-+ router3 +-------|connections
>>    | router2  |      +---------+   |   \in area2
>>    +----------+                    |
>>  
>>  
>>  Hosts discover router3 using Router Discovery; area2 has a different
>>  Pip address, so this makes sense.  Hosts discover router1 and router2

> This sentence confuses me.  Which hosts are discovering router3?
> Those on sn2?

All hosts in area1.  (Remember, all the hosts in area1 share the same
Pip address, and have packets disambiguated by Pip ID.)

>>  using HIP (the Host Identification Protocol); they treat router3

> Why don't hosts discover routers 1 and 2 using router discovery as
> well?

Because router1 and router2 are the same boat as all the hosts.  Also
because the idea behind this is to make it semi-invisible to the hosts
whether they are actually sending the packet to a router or to the
destination host, so long as the destination is within the area.

>>  (which they learned about through RD) just as if it were another host,
>>  for the purpose of directing packets toward it.  Routers discover
>>  hosts in the same way that hosts discover other hosts: they send the
>>  first packet multicast, and then wait for a ``you-found-me'' reply to
>>  come back.

> Huh?  But, a router must "discover" a host before it has packets
> to send to it....so that it can advertise it in routing updates.....

I've been thinking about this, any you may be right, but not for the
stated reason.

My original idea was this:

A host on subnet1 wants to send a packet to a host on subnet2.  It
doesn't know the destination host's SNPA, so it sends a multicast to
the link-layer ``unknown SNPA'' multicast address.  The routers
forward this packet on the other subnets, and so on until a ``you
found me'' comes back, which they modify appropriately and send back
to the original sender.

There is a problem with this, however: HIP packets are regular Pip
packets, so the destination host will send its you-found-me back to
the /original/ host, and we have a potential loop here.  Even worse,
it would mean that the routers would have to wiretap the HIP protocol
to check for messages that might need to be redirected and modified.

So I'll probably change it back to my original original idea, which is
to have the hosts sending out periodic hello packets as well as the
routers, so that the routers all know where the host is, and can
formulate an appropriate you-found-me response themselves.

I've got to start writing this up...

-GAWollman