[v6ops] Trajectory for "transition" mechanisms in our working group

[Fred Baker <[e-mail]>]

On Sep 26, 2018, at 10:27 AM, Fred Baker <[e-mail]> wrote:
> As IPv4 winds down, yes, HE as described will become anachronistic - there won't be a choice to make. It will become, as NAT64 will, a service that is more work to remove than it's worth, but eventually isn't implemented in new implementations and nobody notices.

This time I'm speaking as a co-chair of the working group. This is in part a question: if folks don't see things the way I do, I, and we, need to understand the various viewpoints. But it's also a statement of what I expect the next few years to look like.


As to the status of IPv6, let me comment on its history. I saw an article the other day whose headline announced that "IPv6 has failed; let's pick something else." RFC 2460 was published in 1998, but it didn't define IPv6 as a usable protocol. It needed IPv6-in-DNS, IPv6-in-DHCP, IPv6-in-your-favorite-routing-protocol, vendor implementation, working vendor implementation (oh, is that important?), random features like uRPF etc, ROAs and IRRs, and a long list of whatever else. ICANN didn't approve a policy for distribution of prefixes to RIRs, and the RIRs to networks, until 2006, per OECD. In 2007, about 350 IPv6 prefixes were allocated by ARIN for testing purposes, and in 2009-2011 there was a "run on the bank" in Asia-Pacific as the result of a JPNIC study indicating that IPv4 depletion was a realistic scenario.

In other words, the development of the technologies implied by IPv6 took almost as long to develop as the technologies required to use IPv4 (and involved a few mis-steps). However, the impetus to use them was much lower - whenever a technology was unavailable to make IPv6 work, "well, you can always use IPv4". IPv6 didn't become "important" until "always" became untrue. That started in 2011, when the IPv4 address space available from the IANA (and eventually the RIRs) was depleted; it is becoming increasingly untrue as the cost of purchasing an IPv4 address spirals upward.

"Pick something else" - I have to wonder what "something else" might be in view. We have a number of research projects on the table, such as ICN/CCN/NBN, and they likely have their uses. But I don't see the architectural or business implications following through; if they were rationally available, they would have started a decade back, I suspect. So I see IPv6 deployment as the only rational option on the table.


What I am increasingly hearing from ISPs and other networks is that it is less expensive to deploy an IPv6-only network with (or without) an IPv4 overlay  than to deploy a dual stack network. There are a few networks that have drawn that conclusion; as more and more do, IPv4 will become less of a requirement.

I'm not sure what to expect next, and maybe people can inform me. One option would be two networks that implement IPv6 natively and IPv4 as an overlay service connect, meaning that the address of a 464XLAT PLAT (or whatever) from one carrier is an address in a different carrier for some part of one of their networks. Or, maybe, a network simply stops providing an IPv4 PLAT or other support for IPv4 as an overlay. Maybe they are running lw4o6 and it becomes necessary to do IPv4 routing inside that overlay, so that they want to charge for it as a service. The obviously implied next question is "when" - I'm guessing early 2020s or something like that.


So, what do I expect the next few years to look like? I see three broad classes of ISPs, IXPs, and networks at the edge: those that see IPv6 as the least expensive endpoint and drive to that end, those that have enough IPv4 for the moment and aren't thinking any further ahead than that, and those that are balanced on a knife's edge in between. We have those now. I expect to see a shifting balance, with different markets adopting different perspectives and changing over time. I see a slow but definitely discernible motion from "IPv4 works" to "https://www.youtube.com/watch?v=_y36fG2Oba0" "I suppose we ought to give it a try". One at a time, I hear operators say "I don't want to buy any more IPv4 addresses". The networks that are fixed in the "IPv4 laggard" camp tend to be networks that aren't faced with IPv4 address purchases and ask for a business justification that doesn't involve a monetary outlay.

Ron, Lee, and I asked for a change to the v6ops charter to discuss IPv6-only networking with that in view. What I want to see in v6ops is more of that - lets move IPv4 to an overlay service, and watch it become less relevant over time.

So I personally expect transition technologies of various kinds - lw4o6, 464xlat, MAP-E, MAP-T, and so on - to remain important, but I expect that they will become the aftermarket more than the market. If cloud services are accessible by IPv6, I expect residential access to become IPv6 service to become larger and IPv4 to become smaller. If you want a picture of that, look at Trinidad and Tobago. As I understand it, there are some services (Google, Facebook, etc) that only offer IPv6 access there, so the "IPv6" picture is a picture of their market dominance. https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=TT

I expect service providers to do as little with IPv4 as they can get away with, and for IPv4 to increasingly become ragged around the edges. There are still a variety of places where I would get tarred and feathered and sent home wearing a barrel for saying that, but the SNA people were of that opinion too - until APPN made SNA networks irrelevant and people started deploying IP-native applications.
--------------------------------------------------------------------------------
Victorious warriors win first and then go to war,
Defeated warriors go to war first and then seek to win.
     Sun Tzu