Ivan Popov <pin_at_medic.chalmers.se> writes:

> I can not imagine "tcp connection migration" :)

I've got a lively imagination.  :-)

> The main difference between stable and volatile ip numbers is that
> we'd need a connection context which would be independent of the
> actual ip information, and some heuristics for trying to keep a
> connection, say retrying different local and foreign addresses, with
> the same connection context.  Unfortunately, it looks like a bigger
> change. Anyway, I suspect it will be done, sooner or later.

Intuitively, I'd say there's no problem with servers having fixed IP
addresses -- it might even be a good thing.  However, since the client
is supposed to be "loose", and work with only intermittent access to
the network, it makes sense to let its unique identification to the
server be independent of its IP address.  Venus would probably need to
tie into a suitable interface/routing change notification service in
the kernel (I assume UNIXes in general have these today), in order to
drop and recreate its UDP socket when needed.  The server would just
respond to wherever each request packet happened to come from.

Looking closely at my running Venus, the problem shows up when the
interface it's using to transmit goes away -- it then fails to
transmit at all, because it's trying to use an inactive interface.

If I have it working while on wireless, and then connect the 100Mbps
cable, I get a better route to the server, which is then used for
transmitting, but the replies from the server arrive over the radio
still.  If the actively used interface goes down, however, Venus goes
into a disconnected state.  Reconnecting the previously active
interface causes it to work again, so it's been trying all along to
use the downed interface.

I guess "Mobile IP" is the solution I really want, though, since it
would fix things like ssh at the same time.  :-)

-tih
-- 
Tom Ivar Helbekkmo, Senior System Administrator, EUnet Norway
www.eunet.no  T: +47-22092958 M: +47-93013940 F: +47-22092901