[e2e] the evolution of deployability
J. Noel Chiappa
jnc at ginger.lcs.mit.edu
Tue Dec 17 12:29:49 PST 2002
> From: "David P. Reed" <dpreed at reed.com>
>> the reasons why IPv6 has done poorly, and NAT well, are much more
>> complex than your .. analysis would indicate.
> I wasn't doing an analysis.
Oh, balderdash. There's no great stone book where you go look to read what G-d
wrote as the definitive statement on why IPv6 had problems. Anyone who says
anything about it is by definition doing an analysis; their take, in other
words. And the proof that it's a take is that mine is quite different from
> I have direct testimony from one of the companies about their choice
> to market NAT rather than accelerating v6 deployment, and there is
> public documentation about the other's choice rationale.
So? Just because they made choice X, that doesn't mean that's the main
reason IPv6 had problems. If so, I'm going to annouce that I'm the Queen of
Sheba, and will reality please take note and follow along...
And they might also have decided that the customers were going to pick NAT,
and they could either go with the flow, or get run over... Maybe the
causality in the linkage ran the other way than the direction you postulate?
>> I.e. if you install IPv6 on all your computers - you still need a
>> NAT-type box to talk to the rest of the existing IPv4 Internet.
> Exactly so. So why not sell an IPv6 box instead of a NAT. It's simple,
> it enables more of a future than NAT, and it does the same thing NAT
The big difference here, one you conveniently gloss over, is that with an
IPv6-NAT, you need to install IPv6 on all your hosts. Plug in a v4NAT, and
you're done without touching any of your hosts.
That alone was enough for NAT to kill IPv6.
> What users wanted was home networking with lots of computers on their
> home LAN. Many hated NAT, but that's all they got.
But not from C and M.
Most people running NAT boxes at home are running "Linksys, DLink, Netgear
and similar consumer-oriented vendors", as Henning Schulzrinne points out.
I have an SMC box, myself.
Try walking into your local CompUSA and seeing what they've got out on the
shelf. It's not C. I remember Ran Atkinson telling me, about 4-5 years ago,
that CompUSA was stocking a NATbox/Ether-hub for around $100, and both of us
saying "yeah, NAT's really going to take off now".
Anyway, the fact that vast majority of NAT boxes sold aren't from C and M
kind of voids your thesis, I think.
> IETF had one that could have been rolled out in an early form, but
> instead they had to buy NATs even though they didn't work well and
> were hard to set up.
If they're so hard to set up, why are they flying off the shelves to Joe
Average American from CompUSA?
> Did IP when it originally rolled out have all of its problems solved?
> Nope. Most were deferred.
A small network populated mostly by engineers is a whole different
And I'd also like to point out that deploying something before we had all
the problems worked out was what got us where we are today.
> There were ZERO problems with IPv6 rollout in 1995. It was
> functionally upward compatible with IPv4, and could have been run as
> an overlay on the entire existing IPv4 net.
I'll replay for you my favourite IPv6 aphorism, one you have enough
knowlegde of the technical history of networking to really understand:
"If the designers of IPv6 had been sent off in 1975 to come up with the
next generation of networking, they'd have come back with X.25 with a
bigger sequence number space."
I personally am not too sure what the next generation of networking is going
to look like, but one thing I *can* pretty much guarantee you that it's not
going to be, is IPv4 with a few more bits.
As to why IPv6 had problems, I don't think there's a single reason. There were
a number of factors, most of which we haven't touched on. The poor
initial-benefit/cost ratio was a big one, perhaps the biggest.
More information about the end2end-interest