[LINK] IPv6

[Glen Turner]

Damien Miller wrote:
> 
> > > Mobile Internet devices have been the suggested "killer app" for IPv6.
> > > There are likely to be huge numbers of them which want some level of
> > > unique addressing (though one could argue whether this needs to occur
> > > at the network level).
> >
> > The addressing has to be done at the network layer.  Otherwise
> > every network switch is going to have to track every attached
> > host.  That is, if 3Com sell 100m wireless PalmPilots then every
> > switch in the world is going to need another 762MB of memory [1].
> 
> No - addressing can happen at the application layer too. SMTP is an
> example of this which scales pretty well.

Using application addresses to address mobile devices is problematic.
Their allocation has no mapping to the network topology, and thus
switches are still forced to track the route for every address
rather than track the routes for address prefixes.  All that
substituting e-mail addresses for OUI addressing does
is to use more storage and exhaust the switch's memory sooner.

In short, if you want to have a scalable connectionless network
then you need to keep the size of the routing table within
technological and financial bounds.  This can be done by
limiting the number of hosts, or by aggregating information
about hosts:  that is, using an address heirarchy that has some
mapping to the network topology.

SMTP addressing scales because the Domain Name System 
scales and these names resolve to network addresses
which scale.  Nothing about the e-mail address itself
is inherently scalable.

E-mail also gains from having a percieved level of
'good' performance that allows a disk to be accessed.
You wouldn't want a disk access for every 1.5KB of
forwarded packet.

Finally, e-mail addresses have practical problems that
would limit their performance: they are variable length
(making hardware implementations tricky), have low
information density (so some index or hashing is needed),
and can be very long (255 characters, or 4 addresses
per kilobyte, this forces the use of cheap megabytes of
memory).  The upshot of all these being that a modern
IPv4/6 router will handily forward a packet in the time
it takes for an SMTP-addressed router to read the
maximal-length address.

As this is rather technical, I'd suggest that any further
correspondence happen off-list.

Glen

-- 
 Glen Turner                                 Network Engineer
 (08) 8303 3936      Australian Academic and Research Network
 glen.turner@aarnet.edu.au          http://www.aarnet.edu.au/
--
 The revolution will not be televised, it will be digitised