[NANOG] fair warning: less than 1000 days left to IPv4

Sun May 4 19:08:31 UTC 2008

> I'm not sure that I would tar everyone who does NXDOMAIN remapping with
> the same brush as SPAM and DDOS. Handled the way OpenDNS does, on an
> opt-in basis, it's a "good thing" IMO.

i agree, and i'm on record as saying that since opendns doesn't affect the
people who do not knowingly sign up for it, and that it's free even to folks
who opt out of the remapping, it is not an example of inappropriate trust
monetization (as it would be if your hotel or ISP did it do you without your
consent, or, offered you no alternative, or, offered you no opt-out.)

> I would also say that disaggregating and remarketing dark address space,
> assuming it's handled above board and in a way that doesn't break the
> 'net, could be a "very good thing".

that's a "very big if".

> The routing prefix problem, OTOH, is an artificial shortage caused by
> (mostly one) commercial entities maximizing their bottom line by
> producing products that were obviously underpowered at the time they
> were designed, so as to minimize component costs, and ensure users
> upgraded due to planned obsolescence.

i completely disagree, but, assuming you were right, what do you propose do
do about it, or propose that we all do about it, to avoid having it lead
to some kind of global meltdown if new prefixes start appearing "too fast"?

> Can you give me a good technical reason, in this day of 128 bit network
> processors that can handle 10GigE, why remapping the entire IPv4 address
> space into /27s and propagating all the prefixes is a real engineering
> problem? Especially if those end-points are relatively stable as to
> connectivity, the allocations are non-portable, and you aggregate.

you almost had me there.  i was going to quote some stuff i remember tony li
saying about routing physics at the denver ARIN meeting, and i was going to
explain three year depreciation cycles, global footprints, training, release
trains, and some graph theory stuff like number of edges, number of nodes,
size of edge, natural instability.  couldn't been fun, especially since many
people on this mailing list know the topic better than i do and we could've
gone all week with folks correcting eachother in the ways they corrected me.

but the endpoints aren't "stable" at all, not even "relatively."  and the
allocations are naturally "portable".  and "aggregation" won't be occurring.
so, rather than answer your "technical reason" question, i'll say, we're in
a same planet different worlds scenario here.  we don't share assumptions
that would make a joint knowledge quest fruitful.

> How is fork-lifting the existing garbage for better IPv4 routers any
> worse than migrating to IPv6? At least with an IPv4 infrastructure
> overhaul, it's relatively transparent to the end user. It's not
> either/or anyway. Ideally you would have an IPv6 capable router that
> could do IPv4 without being babied as to prefix table size or update
> rate.

forklifting in routers that can speak ipv6 means that when we're done, the
new best-known limiting factor to internet growth will be something other
than the size of the address space.  and noting that the lesser-known factor
that's actually much more real and much more important is number of prefixes,
there is some hope that the resulting ipv6 table won't have quite as much
nearly-pure crap in it as the current ipv4 has.  eventually we will of course
fill it with TE, but by the time that can happen, routing physics will have
improved some.  my hope is that by the time a midlevel third tier multihomed
ISP needs a dozen two-megaroute dual stack 500Gbit/sec routers to keep up
with other people's TE routes, then, such things will be available on e-bay.

everything about IP is transparent to the end user.  they just want to click
on stuff and get action at a distance.  dual stack ipv4/ipv6 does that pretty
well already, for those running macos, vista, linux, or bsd, whose providers
and SOHO boxes are offering dual-stack.  there's reason to expect that end
users will continue to neither know nor care what kind of IP they are using,
whether ipv6 takes off, or doesn't.

> IPv4 has enough addresses for every computer on Earth, and then some.

if only we didn't need IP addresses for every coffee cup, light switch,
door knob, power outlet, TV remote control, cell phone, and so on, then we
could almost certainly live with IPv4 and NAT.  however, i'd like to stay
on track toward digitizing everything, wiring most stuff, unwiring the rest,
and otherwise making a true internet of everything in the real world, and
not just the world's computers.

> That having been said, I think going to IPv6 has a lot of other benefits
> that make it worthwhile.

me too.