On Thu, 7 Sep 2000, Sean Doran wrote:

# What sucks is the fact that you cannot distinguish what from where
# in either IPv4 or IPv6.

Did I miss something?  (Bill, gimme another 'what' point...).

How do you mean this?  Sure you can.  Don't IP headers carry

host-from/port-from host-to/port-to

information somewhere in the first chunk of space?

Isn't host-from a where, and port-from a what?


# But see http://www.irtf.org/charters/namespace.html
# which is a better solution than either NAT etc. or IPv6.

Hm.  Have to do that when awake.

You cannot simply compress net addresses without losing something
somewhere.  If you don't have the IP addresses sufficient to negotiate
the namespace/address space, something's got to give.  Since the port
and the protocol are the only other things in the header (outside of
the MAC address and the length, either of which will screw you if
they don't match right), they're the only parameters which can be
dinked with in order to provide a wide enough space of addressing,
and I somehow don't think you want to be mucking with the protocol number
as a means of address expansion :-).

So, okay, we can do ports between 1024 and 65533 (to be "safe").  That
means that one IP address can do several class C mappings, but that
limits the throughput, port-wise, quite severely.  Imagine what happens
when things get out of sync and Don't you DARE say "That can't happen"
because it has.

Not to mention, as has been pointed out, forcing protocols to wrap
themselves around a desperate measure instead of being allowed to
depend upon a pre-existing specification is sheer lunacy.  If it works,
fine, but don't count/insist on it.

At this point I welcome anything that can relieve the pressure of IPv4
congestion, but I think that IPv6 has more to show, so far, than just
about anything else.  They're actually playing with it and implementing
it, testing and in sort-of real-world situations.  I'd like to see it come
to fruition, myself.

				--*greywolf;
--
BSD: the second best thing you can get for free