In message <m3y9fvygdf.fsf@test9.crm.mot.com>, Alexandru Petrescu writes:
> Alex Zinin <azinin@nexsi.com> writes:
> > Specifically, one could envision an architecture, where mobility is
> > absolutely transparent to the routing system by making address be
> > always topologically signi- ficant and having a separate mechanism
> > for name-to-address mapping...
>
> Following this line, is it possible to imagine an architecture where
> addresses are topologically significant and prefixes are also
> topologically significant? And then having a mechanism for
> name-to-prefix mapping?
>
> I might be far off from what "topologically significant" means
> exactly.
>
> Alex
IP Mobility allows a fixed addressed entity to move and reports its
address "home" (oversimplified). A fixed loopback address and plain
old routing would do the trick for mobility if it weren't for the fact
that a /32 would have to be advertised for every mobile entitity. To
avoid that traffic is relayed through the home station.
Mobility seems to be a can-of-worms because people can't seem to get
their heads around the idea that without changing hosts, you either
have the routing inefficiency of carrying around very specific routes,
/32 for IPv4, or the forwarding efficiency of using a home station for
global routing and then encapsulating and forwarding.
If you do accept that changing hosts is fine (every host in the
Internet, not just the mobile hosts) then you can have DNS map some
DNS record (or whatever you revolutionaries decide DNS should be
replaced with) that replaces an A record to information to which a
query for the current address can be launched. When the the address
changes (ie: suspend laptop, get off flight, open laptop and get new
IP address) all other ends of open TCP connections can be notified of
the change. Given a ICMP reply the other end can repeat the query to
see if the mobile host moved (in case both are mobile) before giving
up on a TCP session. If DHCP leases last longer than TCP connection
timeouts, then there should be no RST from a new lease holder after a
mobile host gives up an address and moves elsewhere.
Here we are no longer talking about just changing interdomain routing
or routing for that matter. This might be a nice followup after the
IPv6 migration is completed, or maybe rolled into the IPv6 migration
(which is making very slow progress). Changing all host
implementations is no small matter.
btw- Note that the side of the TCP connection that does the listen and
accept does not do any name lookups, therefore it has no name with
which to repeat a query and find a moved mobile host. It must be
notified of the move. The connect side of a TCP connection can repeat
a query if the other side moved. Of course, I'm assuming changing TCP
to accommodate new routing requirements is still off the table.
I'd have to say that universal deployment of intercepting intermeiates
to support mobility is even less likely to occur than host changes. I
also think neither is at all likely to change to support mobility.
Curtis
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:04 EDT