Update to BCP-38?

Thu Oct 3 15:22:59 UTC 2019

On Oct 3, 2019, at 9:51 AM, Stephen Satchell <list at satchell.net> wrote:
> It appears that the only parallel paper for IPv6 is
> draft-ietf-v6ops-ipv6rtr-reqs-04, _Requirements for IPv6 Routers_, which
> currently carries a copyright of 2018.  It's a shame that this document
> is still in limbo; witness this quote:  "It is inappropriate to use
> Internet-Drafts as reference material or to cite them other than as
> 'work in progress.'

Speaking as v6ops chair and the editor of record for 1812. draft-ietf-v6ops-ipv6rtr-reqs kind of fell apart; it was intended to be an 1812-like document and adopted as such, but many of the "requirements" that came out of it were specific to the author's operation and not common to other operators. So it ultimately didn't happen.

> Someone else mentioned that "IPv6 has been around for 25 years, and why
> is it taking so long for everyone to adopt it?"  I present as evidence
> the lack of a formally-released requirements RFC for IPv6.  It suggests
> that the "science" of IPv6 is not "settled" yet.  That puts the
> deployment of IPv6 in the category of "experiment" and not "production".
> 
> Is that really true?

That's a long story. The IETF realized it needed a next generation protocol in 1990; that's where NATs came from, the successive efforts to recover unused IPv4 space, and research into possible next generation protocols. IPv6 was proposed in 1993-1994, originally published in 1996 as RFC 1883, and republished in 1998 as RFC 2460. It was recently re-re-published as RFC 8200.

Supporting work was required in DHCP, DNS, and several routing protocols; that happened over time. ICANN didn't adopt a policy for the allocation of IPv6 address space to RIRs until 2006, and in 2007 there were a number of allocations from IANA to the RIRs and from some of the RIRs to operators in various parts of the world. Testing was also important - primarily done by the NRENs. That wound up with comments going back to various vendors. I was employed by one of them, but will refrain from giving "insider" comments. Suffice it to say that there were multiple vendor implementations, mostly incomplete in one way or another. 

IANA allocated the last five IPv4 /8s to the RIRs in 2011, and since then the IPv4 address market has been mopping up the slop. Per https://ipv4marketgroup.com/ipv4-pricing/ (if addresses were real estate, ipv4marketgroup would be a real estate agent), the price of an address was stable at or near $10/address from several years, and in 2016-2018 shot to about $18/address. They expect the price to start to fall in the next year or so, as CIOs figure out that its a waste of money. 

There is no demand until further IPv4 deployment no longer suffices. I would say that there was no real market demand until after January 2011, and probably 2012 or 2013.

At this point, there is fairly wide deployment among the ISP and CDN operators, and vendor implementations are fairly complete. Google, APNIC, and Akamai report on traffic levels; Google says that they see at least 5% of the requests they receive from 61 countries use IPv6, and from one country a tad more that half of the requests they receive use IPv6. https://www.vyncke.org/ipv6status/compare.php?metric=p&countries=be,yt,de,gr,my,vn,in,gf,us,uy,fr,tw,jp,lu,ch,mx,br,pt,fi,mq,ax,th,ee,re,hu,gb,ca,gp,tt,ie,lk,nz,au,pe,ec,nl,ae,ro,ga,bo,sa,sx,cz,no,si,sg,pl,gt,at,mo,ar,mm,kr,fo,om,lv,zw,pr,ke,tg,ba. And interesting point in those reports is that Google and Akamai are CDNs, which means that (for the most part) a request goes almost directly from a user's broadband interface to the service in question. APNIC is different in that it operates no CDN; requests they receive cross the backbone, and therefore also measure backbone deployment. To that matter, let me list the APNIC charts of the top ten:

https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=BE
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=YT
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=DE
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=GR
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=MY
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=VN
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=IN
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=GF
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=US
https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=UY

In each of those, APNIC measures and reports a distinction between a requestor being "IPv6 Capable" and "Preferring IPv6". This is done using a google ad that includes three one-pixel links; one of the names has only an A record, one has only a AAAA record, and one bas both. If the requestor uses the first two links and APNIC receives the SYN, then the end system and every AS en route used and provided an IPv4 or IPv6 capability respectively. In the third case, the end system presumably gets both an IPv4 and an IPv6 address, and makes a choice. If it chooses IPv6, it is reported as preferring IPv6. If you select the links above, you will find that end systems (telephones, laptops, whatever) will generally prefer IPv6 if given the option.

Also interesting on those pages, APNIC lists the ASNs serving the indicated country, and separates requests by ASN. For example, in India (https://stats.labs.apnic.net/ipv6/CC?x=1&s=1&p=1&w=30&c=IN), APNIC reports on 100 different ASNs, of which half demonstrate non-trivial IPv6 traffic, and (I think) 20 demonstrate more than 10%. A celebrated one among them is Reliance JIO; https://stats.labs.apnic.net/ipv6/AS55836?c=IN&p=1&v=1&w=30&x=1. You might look at that chart, as it breaks out Reliance JIO and "rest of country". Reliance was a new operator in 2011, and as I understand the story got a small allocation from APNIC (in accordance with its policies) and went to the market to purchase IPv4 address space. They recognized the expense of that approach in perhaps 2015, and started IPv6 deployment. At that time, they were probably the only, or at least the principal, operator using IPv6 AT ALL in that country. What the data shows is that other operators - half of those in India - have since then followed suit, probably due in part to competitive pressure.

So we see IPv6 in broadband, in ISPs, and in telephone networks. To give you an anecdote, my kids have teased me about IPv6 for years, and each now have IPv6 service from their various ISPs and telephone networks despite themselves - and use it for the majority of their accesses.

What is visibly lacking is enterprise deployment. There are companies that have it, but they are unusual; those that are visible in routing usually have it for their outward-facing services such as mail and web, but not internally. To be very honest, some of that leaves me cross-eyed. A recent example I was asked to give advice on - name consciously left out - was a country that wanted to deploy an NREN that would enable certain services to some 5000 secondary and university-level schools. If they deployed it using IPv4, simply buying the addresses would have cost them $25M; equipment and personnel are not included in that expense. From their RIR, getting the smallest allocation the RIRs give to an ISP - an IPv6 /32 - would cost $2100/month., and would give them allocations for 65535 such end sites. So they have a choice: $25M up front, or $2100/month. Hmm, that's a hard one... MIT recently sold half of their IPv4 address space - 1 /9 out of a /8 - to Amazon, and stated in a blog that they planned to use the proceeds for their IPv6 deployment. A company that I am on the board of independently from me decided to sell half of its legacy /16, and use the proceeds for internal projects. If I were an enterprise CIO, I would see that address space as banked money, and want to access it. I'm obviously clueless - that's not happening.

And on lists like this, I am told that there is no deployment - that nobody wants it, and anyone that disagrees with that assessment has lost his or her mind. That all leaves me wondering which of us doesn't quite have their eye on the ball.