[arin-ppml] The non-deployment of IPv6

[Michel Py]

Fred,

I am a bit surprised you swallowed that bait; the IPv6 camp should be
desperate to send people of your caliber feeding controversies from me.
But I'll bite! I changed the subject; "IPv6 is a failure" is too
provocative and it is too early to make this determination yet. I hope
you will agree that the current state of IPv6 is "non-deployment".


>> Michel Py wrote: "IPv6 is a failure" ..[snip].. and a statement
>> that would have been widely seen as politically incorrect a few
>> years ago suddenly becomes an uncomfortable, unwanted but
>> nevertheless more prevalent every day reality check.

> Fred Baker wrote:
> I agree that it is an important reality check. That said, I
> think that IPv6 has been since 1996 a solution developed in
> expectation of a problem.

Unfortunately, only for some parts. This statement is valid for the
larger address space solving the expected problem of IPv4 shortage (and
to a lesser extent the problem of very large ISPs being too big for
10.0.0.0/8 and even larger-than-class-A spaces). Some other parts of
IPv6 are either a solution looking for a problem, an undelivered
promise, or vendors trying to have it their way despite what the
consumer base says.

- Solution looking for a problem: let's look at autoconfiguration:
first, it's not new to IPv6; we had that with IPX. Second, it addresses
a problem that ceased to be one for a long time: DHCP is working and is
available on any $20 cheesy "router". Third, it's not much of a
solution: if it was, there would be no need for DHCPv6.

- Undelivered promise: routing table aggregation. If any IPv6
multihoming solution (except like doing exactly the same as IPv4) was
working, the RIRs would not have already adopted policies allocating PI
addresses to non-LIR entities.

- Vendors trying to have it their way: I understand this is not the
forum to debate my pet peeves with Cisco, but I will use a recent
real-world example:
Two weeks ago, I deployed a new customer network. Small; a handful of
remote sites. The routing protocol I used (and it was my call to make)
is EIGRP, because RIP and IGRP sucks and OSPF is too complex for the
customer. Where's EIGRP for IPv6? Maybe you could hint someone at Cisco
(evil grin) that instead of IPv6 with OSPFv6, the customers will keep
running IPv4 with EIGRP? 


Unfortunately, the IPv6 marketing is still using outdated arguments. The
bottom line is: the only real problem that IPv6 solves today is the
predicted shortage of IPv4 addresses. Some aspects of IPv6 are indeed
superior to IPv4, but not worth the increased complexity and cost. And
we mostly agree (see below) that in the end it's just a matter of big
bucks.


> People generally don't do things that cost them money until they see
> something resembling ROI, and folks on this list and in other places
> have seriously questioned the ROI.

With some good reasons.


> In my opinion, IPv6 will have been demonstrated a failure if (a) the
> IPv4 address space doesn't run out or (b) when it does, IPv6 turns out
> to not be an adequate solution. It will have been an adequate solution
> perhaps *just* adequate, but adequate) if it gets deployed widely and
> as a result it becomes more straightforward for operators (ISP and
> enterprise) to run their networks. It will have been inadequate if in
> the long run if we see sustained use of complex work-arounds - 6to4,
> Teredo, ISATAP, 6rd, ds-lite, IPv4/IPv6 CGN, IPv4/IPv4 CGN, IPv4 A+P
> hacks, and so on - instead of IPv6 deployment.

I agree with what's above. Unfortunately, there are so many workarounds
already that nobody really believes that they will be quickly eliminated
should native IPv6 deployment occur. We all have in our respective
closets a few skeletons of temporary-only-for-3-days-I-swear ugly hacks
that are still there 5 or 10 years after the fact. In other words: we
don't even have a clean start, and a large number of unproven hacks does
not appear any better than double NAT.


> The holy grail of minimized opex/capex to be found in relative
> operational simplicity, and that is IMHO to be found in a uniform
> contiguous address space.

I agree with that too, but IPv6 brings a whole lot of new complexities,
one of the major no-nos being multiple addresses per host. Talk about
operational simplicity. Between two evils none of which is clearly a
lesser one, people will keep with the one currently in place.


> In my mind, IPv6 is the only approach on the
> table that has a hope of reducing costs.

So far, it seems that not too many new people are believing in this
hope, and that more of those who used to believe it have become
disconnected from their previous faith.

As of being the only solution on the table, it is. Double-NAT is
temporary-only-for-3-days-I-swear, right?

I would entertain that IPv4+ (which would be a backwards-compatible IPv4
with the only difference being an extended address space) would be much
more popular as a solution if it was on the table.

Question: how many years after the last IPv4 block gets allocated to the
RIRs do we wait until we feel it is time to make that call?

Michel.