On Tue, Aug 11, 1998 at 09:19:02AM +0200, Ignatios Souvatzis wrote:
> Yes, but what would be the win? (Other than maybe a cleaner interface in

One win would be imho, that you need no addresses on p2p interfaces
which saves IP addresses.

The other is merly a theoretical construct:

  a------p2p link-------b--------c
  |                     |
  +----ethernet---------+

With the current scheme, you cant route a<->b over ethernet and a-->c
over the p2p link.

The other thing that I have in mind are encapsulating tunnels.


  ---> =default route from a
a------b-------c------d--------e
|                     |
+------tunnel---------+

you define e.g.

ifconfig gre1 a d netmask 0xffffffff
and set the tunnel source to a and the tunnel end to d.

Now with route over the tunnel  (e.g. route add e d) 
a packet gets delivered to the gre1 interface, where it is
encapsulated in a ip-packet that has src=a and dst=d. This one gets
delivered from ip_output to gre1 and so on.

With interface routes you would say
ifconfig gre1 a
and set the tunnel source to a and the tunnel end to d.
and then route add e gre1

here there is no specific (host) route to d inserted in the routing
table and the encapsulated packets would just flow over the default
route to d where it would be decapsulated and delivered.

-- 
          See <a href="http://www.netbsd.org">NetBSD</a> for a multiplatform OS
echo '16i[APq]sa427A902E563D30E[dd0=a35%20+P35/lbx]dsbx'|dc
							-- Adam B. Roach