Re: new version of draft-narten-ipv6-3177bis-48boundary-03.txt

[james woodyatt <jhw@apple.com>]

On Dec 6, 2007, at 09:24, Gert Doering wrote:
> (I do wonder a bit why zero-conf devices would all of a sudden  
> invent lots of independent *subnets*, though).

Consider how RTMP worked in Appletalk (except, try not to get hung up  
about the historical problem of excessive unsolicited broadcasts—  
pretend it was designed by people who understand how to do  
exponential backoff and keep multicast traffic to a minimum).

Zero-configuration IPv6 routers would need to generate random subnet  
identifiers from a sufficiently large space to ensure that collision  
is statistically rare, because collision requires renumbering, which  
causes more traffic.  When the subnet identifier space is large  
enough, then the network stabilizes quickly after a collision, i.e.  
typically after only one iteration.  If the subnet identifier space  
is too small, then network add/join events might require many  
iterations of the resolution protocol before an available subnet  
number is found.

Why might zeroconf routers be more prevalent in the future than they  
are now?  Answers: p1) IPv6 forwarding is cheaper and easier to do in  
hardware than IPv4 forwarding, so L3-switches should become cheaper;  
p2) the day might never come that IEEE 802.1 is the only sub-IP layer  
one ever needs and bridging (not routing) is how networks might  
therefore be interconnected in residential environments; and p3)  
consumer devices are rarely configured for any mode other than the  
factory default, and developers need to choose defaults that are  
optimized for the broadest possible set of usage scenarios, i.e.  
secure out of the shipping container, with a firewall turned on until  
the user encounters a need to reconfigure or disable it.


--
james woodyatt <jhw@apple.com>
member of technical staff, communications engineering