IPv6 deployment excuses

Mon Jul 4 14:10:47 UTC 2016

On Mon, Jul 4, 2016 at 7:44 AM, Matt Hoppes <
mattlists at rivervalleyinternet.net> wrote:

> I disagree. Any data center or hosting provider is going to continue to
> offer IPv4 lest they island themselves from subscribers who have IPv4 only
> - which no data center is going to do.
>
> Thus, as an ISP you can safely continue to run IPv4. Ipv4 won't be going
> away for at least ten years or more - if ever.
>

That's an interesting "bet the business" decision for an ISP to make. It's
not one a large ISP can make simply because they want to continue growing
their subscriber base and that's a losing proposition as far as profits go.
That's why all the big ISPs are either implementing IPv6 or actively
working to deploy it. So it seems you're saying that a small to medium
sized ISP can delay 10 or more years because all the content their
customers want will be available over IPv4.

And that's pretty much betting the entire business on what is basically
nothing more than a crystal ball. I don't know about you, but I think back
to the mid to late 80s and most ideas I saw about where technology would be
by the mid to late 90s were pretty inaccurate. Jump to the mid 2000s and
past projections were pretty off-base again. Shortly after that, mobile
computing really hit and the world today looks very little like it did
then. Do you really think someone should bet their entire business on your
ability to make Internet forecasts 10 years into the future?

Now, that's not to say I expect IPv4 to necessarily be mostly gone (from
the Internet, not private networks) in 10 years, though it wouldn't
particularly shock me either if things did work out that way. But I do
expect a tipping point will be reached well before then that reduces the
utility of IPv4. Technology changes on the Internet have not traditionally
been steady, gradual processes. Rather, they've had some sort of fairly
long lead time, a rapid spike of uptake, and then a flip from a 'new'
technology to something expected. There's then often something of a long
tail, but it can be pretty unpleasant to be forced to exist in that tail.
The attitude quickly switches to one of, "Oh, you're still using 'x'? You
should use 'y' instead. It's working fine." And issues with 'x' get lower
priority attention. And that 'flip', when it happens, tends to happy
relatively quickly.

Often, it can be difficult to predict if a new technology will overtake and
supplant an older one. The IPv6 transition, however, is being forced by
IPv4 exhaustion. That's putting a lot of technological and financial
pressure on most of the parties involved. As someone who works at an
enterprise that sees a lot of traffic, primarily from the US, we were
seeing a steady increase in IPv6 traffic from end users from practically
nothing in 2012 to around 15% in 2015. This year we've seen it spike to
25%-30%. So in the US, we may very well reach that tipping point within the
next couple of years. If we do, the utility of IPv4 will probably start to
degrade pretty rapidly as more attention and focus is placed on IPv6
connectivity. If that happens and you're still an IPv4 only edge/access
network that hasn't even begun planning an IPv6 deployment? That's apt to
be an uncomfortable experience.

But again, I'm not a prognosticator. I wouldn't have correctly guessed the
timing for any of the transitions I've seen in the past, though I did
sometimes come close to guessing the outcome. (That's one of the reasons I
started a small scale production deployment of Linux at my place of
employment back in the mid-90s, something we now have running on platforms
all the way up to our mainframes.) It looks to me like, in the US at least,
we're on the 'rapid uptake' slope of adoption. If that's the case, then
that tipping point is probably coming a lot sooner than 10 years out. You
could be right and everything will be fine for IPv4-only customers and
networks in 10 years. But that is most definitely a high stakes bet to
make. I certainly wouldn't be willing to make such a gamble.

I also want to note that enterprise or data center networks moving to IPv6
only does not necessarily involve NAT64 or any sort of translation. For any
large internet service, inbound connections are typically terminated at the
edge. A new connection is then established from the point of termination to
the data center resources. So Facebook, for instance, only needs to
dual-stack its edge. And if you use a 3rd part CDN for the edge, you don't
even have to do that. That's what other posters were pointing out.
Depending on its security profile, a large enterprise network might also
'proxy' outbound Internet traffic (primarily web, mail, DNS) already for
its internal users. If that's the case (as it is where I work) very little
outbound translation is required as well and only the outbound perimeter
services need to be dual-stacked long-term. So if an enterprise or data
center network operator isn't already thinking in terms of where they can
go IPv6 only rather than dual-stacked, now is probably when you want to
start thinking that way. There is a definite cost to trying to operate what
is essentially two networks over the long haul. The more places you can
move from dual-stacked to single-stacked IPv6 in your network, the better
off you'll be.

And even ISP access networks are moving more toward offering IPv4 as a
service on top of a native IPv6 network. T-Mobile is already doing that.
>From what I've seen, others are exploring ways to do it. Especially given
all the constraints on IPv4, that approach just makes sense.

So again, the safe bet today looks like IPv6 to me. Wagering that the rest
of the Internet will be fully supporting IPv4 at the same level of utility
as IPv6 in a decade looks like a high-risk gamble.

Scott