[governance] Re: Quo Vadis IPv6

[John Curran]

Karl - 

Wow. I'm not quite certain where to start in responding; my recollection 
is somewhat different on many of your points (both historical and present 
outlook) - I believe I'll constrain myself to those which are germane to 
the question of technical community performance in managing the IPv4/IPv6 
transition, and leave the rest for another time.

Disclaimer:  I was part of these proceedings in many capacities, including 
a member of the IETF IPng (IP next generation) directorate, one of the TUBA
developers, and IETF Operations/Network Mgmt Area Director, and involved in
the early days of the ARIN and ICANN formation.  I'm not saying all of this
to establish credentials as much as to make sure people understand that when
I am critical of some of these outcomes, I often share directly in the blame 
due to varying degrees of involvement in the decision making.

With that preface, let's proceed on a historical retrospective on the
development of IPv6 (which was called at the time "IPng") -

On Jun 16, 2011, at 7:02 PM, Karl Auerbach wrote:

> On 06/16/2011 02:30 AM, Izumi AIZU wrote:
>> Or, what was the biggest reason/rationale not to make IPv6 compatible
>> with IPv4....
> 
> IPv6 had a somewhat difficult birth back in the early 1990's.
> 
> There were actually several proposals - my own favorite was a thing called TUBA, which was an adaptation of the ISO/OSI connectionless network layer.  There were several aspects that were interesting, and it had an address that was expansible up to 160bits.  The hostility towards ISO/OSI is still strong today - much to the detriment of the internet - and was much stronger back then.  So TUBA sank beneath the IETF's waves.

There was actually a true bona-fide competition among all of the next generation 
proposals... each  had to be well-specified and withstand ample discussion by the 
technical community of any potential flaws as well as potential benefits.  Several 
never reached the level of clarity to proceed, others merged with similar next 
generation proposals in an interesting form of inter-proposal fratricide, etc.

> It was recognized back then that there were several issues in play; the address size was recognized as but one issue among many.
> 
> The format of the address was another - the variable size of the TUBA "NSAP" scared people who built routers because of the overhead of parsing a flexible address format.


Agreed - there was significant discussion of 64 bit, 128 bit, and various
variable length addresses schemes and the perceived benefits of each. (FYI
a nice list of historic documents is here: http://rms46.vlsm.org/1/42.html, 
as well as Scott and Allison's book "IPng, Internet Protocol Next Generation"
published Addision-Wesley, 1996)

While some of the proposed IPng candidates achieved running code before
the decision, it should be noted that the focus was on actual base IPng 
protocol operation, not any form of transition or interoperability with 
IPv4.  Specification of transition mechanisms was not necessary to be a
valid IPng candidate, and hence it was always considered future "TBD" 
work.  (When work on IPv6 transition mechanisms finally did appear, 
post-selection, it was focused on enabling early IPng researchers to 
tunnel individual hosts and development networks together to make an 
early "IPv6 Internet" over the existing IPv4 Internet.)

In all of the IPng discussions, there was general acceptance of several
assumptions which in retrospect may have been suboptimal to the design
of overall result:

A) The current Internet (circa 1995) is simply too big to allow changing
  of the installed IPv4 base, i.e. anything we did had to interoperate
  with the systems already installed, as opposed to those which would 
  be installed in the following 15 years...

  Making IPng work with IPv4 is very difficult unless you accept that
  you might need to make minor changes to the future IPv4 installed base
  which is going to be deployed over the coming decade, so as to smooth 
  the introduction of IPng via various compatibility and transition hooks.

B) The way that router hardware worked in 1995 represented the best 
  information that we had, and so we should not do anything that we
  don't know how to build fast hardware for on day one.  This means
  that using existing IPv4 option fields, or doing variable length
  headers were effectively deemed to be too performance-limiting and 
  ruled out accordingly.  

  While it is quite probable that that IPv6 would have picked up a
  "performance" stigma if it had explored these options, we might
  have gained better interoperability with IPv4 and/or enormous 
  flexibility in addressing as a result.

C) The goal is to create a new protocol which might run side-by-side
  of IPv4 initially, but would be intent on displacing IPv4 rather 
  than achieving long-term interoperability with IPv4.  In effect, 
  we were designing IPv6 as we wanted IP to look originally, as if
  it were the only protocol, and not documenting its interaction 
  with the existing IPv4 Internet.  This was a natural side effect
  of the IETF simply working on technology whereas the Internet 
  was the largest installed base of that technology; any standard
  for how the Internet would actually change from IPv4 to IPng was 
  beyond scope of the IETF and was to be determined by individual
  Internet service providers. This is an inherent side-effect of 
  defining the Internet as a collective private-sector initiative;
  none of the IETF, ISOC, IANA, RIRs, or ICANN could require a 
  more coordinated transition plan, as each of these organizations
  only provided certain technologies, standards, or basic services
  which were to be used by service providers to enable them to build 
  the Internet via the interconnected set of service provider offerings.  
  (The closest thing to an actual Internet transition plan was my
  absolutely-non-binding & full-of-disclaimers RFC 5211, which notes 
  in passing that companies that don't speak about their IPv6 plans 
  may be surprised when others expect them to be IPv6 in 2012...)

Despite these assumptions, the IPng effort was a remarkable success, 
in that it resulted in functional stable specification for a new 
version of IP which would fix the primary problem of insufficient 
address size.  Because this was completed in the late 90's, it was 
incorporated in many computer operating systems as well as common 
networking gear.  If we did not have IPv6 already well deployed in 
PC and server operating systems today, it would be inconceivable to 
discuss moving to a new version of IP that required coordinated 
operating system upgrades simultaneously with the network changes.
As an aside, this is also why developing a new, alternative answer 
to the IPv4 depletion problem today isn't really viable; it's taken 
us 10 years to get IPv6 in the common base operating systems that 
are globally deployed; we don't have another 10 years of life in 
IPv4 but that is what it would take to complete the development 
and deployment of any other solution.

> Early on there was much talk and though about IPv6 transition - how things might co-exist, even with intermediated interoperation of IPv4 and IPv6 devices.  But over time the energy to have a smooth transition withered and left us more with a conversion from IPv4 to IPv6 rather than a transition - the difference is subtle, conversion tends to be a more painful hurdle to leap than a transition.

Absolutely correct.  Many in the technical community felt that IPv6 would 
be deployed simply because it was "the right thing to do" and "it would be 
better". I went on record with RFC 1669 (August 1994) indicating that IPng
was not going to be deployed because the IETF pushed it and that lack of 
any meaningful new functionality plus the assured appearance of NAT would 
result in IPv6 being ignored until the last minute.  (FYI - For those who
don't believe that the universe has a sense of humor, recognize that despite
such it is now my fate to be one of the leading advocates for IPv6... ;-)

> My own personal feeling is that IPv6 is too little and too late, that it will hit with about the same force as ISO/OSI - which like IPv6 had the backing of governments (GOSIP) and large companies (MAP - General Motors, TOP - Boeing).

On this point, I completely disagree.  We are indeed going to be in for 
an interesting ride, but there are actual solid signs of a real emerging
production IPv6 version of the Internet, and there is enough activity over
the last 6 months that I believe IPv6 enablement of the majority of the
Internet content will indeed happen over the next 3 to 5 years.  It is not 
100% certain, and it's definitely longer than we all would like, but we 
are now finally seeing real progress and prioritization of these efforts.

FYI,
/John

(speaking on behalf of myself only)



____________________________________________________________
You received this message as a subscriber on the list:
     governance at lists.cpsr.org
To be removed from the list, visit:
     http://www.igcaucus.org/unsubscribing

For all other list information and functions, see:
     http://lists.cpsr.org/lists/info/governance
To edit your profile and to find the IGC's charter, see:
     http://www.igcaucus.org/

Translate this email: http://translate.google.com/translate_t