The YMDD: (Flag Day Format) for the 128-bit DNS could focus on using the,
often un-used, Strict Source Routing[1] feature of IPv4.
[1]
http://www.isi.edu/in-notes/ien/ien156.txt
http://www.google.com/search?hl=en&ie=UTF-8&oe=UTF-8&q=strict+source+routing
http://www.iss.net/security_center/advice/Underground/Hacking/Methods/Technical/Source_Routing/default.htm
http://www.info.fundp.ac.be/~infonet/coursenligne/Info2231/INFO2231-5/img20.htm
==============================================================
As an example, if Sept 11, 2002 was the upcoming flag day, then the 128-bit DNS
could
contain AAAA records with two hops to place in the IPv4 options field and then
a final [IPv4]:[Port]
address.
2911:[IPv4x]:[IPv4y]:[IPv4z]:[Port]
This would say, on 9/11/2002 you can reach [IPv4z]:[Port] by first sending the
packet to [IPv4x]
and then sending the packet to [IPv4y] and finally to [IPv4z]:[Port]. Sixteen
years later, the same prefix
can be reused. Most packets would have made their way through the legacy
Internet by then. Each day,
the owner of [IPv4z]:[Port] could change their location and queue that in
advance. Applications could
assume that the latest date holds and normal time-outs on AAAA records could be
used to refresh any
local DNS caching. If the owner of [IPv4z]:[Port] only connects their PC behind
facilities operated by
the owner of [IPv4y], then, they combine to have a 64-bit address [IPv4yz]. A
typical suburban home,
able to get one static [IPv4y] static address, would then be able to have over
4 billion (4,294,967,296)
people living there. The Top Tier providers could of course exchange the
traffic across a core with all
the [IPv4x] addresses around the edges of that cloud. It should be clear who
owns the [IPv4x] addresses.
http://www.iana.org/assignments/ipv4-address-space
In theory, different people could own the [IPv4y] addresses. Managing that
layer, could be a challenge,
even for the largest MLM - Multi-Level-Marketing companies.
The Flag Year Format (FYF) approach of having 2002:[IPv4x]::[IPv4z]:[Port]
helps to protect the
incumbent [IPv4x] owners and leaves room for more clarity to be added to the
]::[ no-man's land.
The devices placed at the edges of the [IPv4x] get to define what happens in
]::[ "no-man's land".
If all of ]::[ "no-man's land" is given to the end-user, then, they have an
80-bit address space to use.
One dynamic address ([IPv4x])can be used to get their 80-bit address space
routed. Each year, a
new "version" can be defined, with more and more of ]::[ no-man's land defined.
If more bits are taken
from ]::[ no-man's land in the 128-bit DNS to set bits in the IPv4 Header, then
there are less bits for
80-bit end-user addressing. The incumbent [IPv4x] owners do not care, because
they get the traffic
via their lock on the 2002:[IPv4x] version, and influence with large companies
to encapsulate to help
them retain their monopoly. Users are on the other side of ]::[ no-man's land
with their [IPv4z]:[Port]
address trying to get routed. The natural instinct is for the incumbents to
grow to the right and for the
users to grow to the left, and they can meet in the middle. That may not happen
if DNS128 bits are
used to set TOS(8), Identification(16), Don't Fragment(1), TTL(8) and
Protocol(8) bits in the IPv4 Header.
There are only 32 bits in ]::[ no-man's land. Setting all those fields would
require 8+16+1+8+8=41 bits.
One could drop the ability to set the Protocol(8) via DNS and one of the TTL(8)
bits and define
next year's version to
be....2003:[IPv4x]:TOS(8),ID(16),DF(1),TTL(7):[IPv4z]:[Port]....this would
allow for the 128-bit DNS to be used to do late binding on most of the key IPv4
Header values as
well as the UDP|TCP Port value. Two IPv4 Headers would be used, one vanilla one
for global routing
and the other under more control by the end-user, assuming they control the
contents of the DNS AAAA
records. The 64-bit effective addressing then contains a site-id part(x) and
the end-user(z) part.
[IPv4x]:[IPv4z]:[Port] The other bits would be pulled from the 128-bit DNS to
set other values in
the IPv4 Header which are typically not used for routing....TOS Routing changes
that of course.
Routing based on the Identification Field also can change that.
http://www.netfilter.org
Another approach is to split the 32 bits in ]::[ no-man's land into two 16 bit
fields, one for the incumbents
to manage and one for the end-users to manage. If the incumbents and the
end-users end up with the
same approach, then there is symmetry and end-users and the incumbents can all
be equals on the same
end-to-end cloud. If you couple that with the traditional use of DNS for
destination node addressing information,
you quickly see that 16 bits can fit nicely into half of the TOS and
Identification Fields. TOS(8),ID(16)
The Don't Fragment bit DF(1) may not be a desirable bit to control via the DNS
as networks become more
capable of not fragmenting packets enroute. Taking half of the TOS(8) field and
half of the ID(16) field
one ends up with 4+8=12 bits to fit into 16. Another 4-bit value can be
accommodated. Unfortunately,
all 16 bits of the Identification Field can not be used if some legacy
functionality is to be preserved. One
can back that off to 14 bits, which results in half being a 7 bit field. The
7+4=11 combination begins to
emerge as a useful package of extended address bits. The 11 bits fit easily
into the 16 bits of each side
of ]::[ no-man's land, with 5 bits to spare. The extra 4 bits of TOS can be
reused there, leaving one
remaining bit on each side of the fence. Those two bits can be used to finesse
the transition from networks
that support fragmentation to those that do not.
00 - Legacy
01 - Don't Fragment End-User
10 - Don't Fragment Incumbent
11 - Don't Fragment Either
Leaving the 4 TOS control bits out of the picture and the DF bits, one sees
that 11 bits remain
as the most bits that can be promised to IPv4 end-users for extended
addressing. With 11 bits and
2,048 Registries, a more manageable extended address space structure is
possible, compared to
what one might face with 4+ billion registries and [IPv4y] addresses in the SSR
approach. With
the incumbents and the end-users each gaining 11 bits of addressing there is
symmetry and more
fairness. Some endusers of course do not want to give up any of the bits in
]::[ no-man's land. As
a result, the fence currently sits with 2002:[IPv4] isolated, becoming more
crowded, and increasing
in value. It is ironic that those who think they are doing away with the
incumbents are actually
playing into their hands. Moving the fence to the middle, equalizes the playing
field and everyone wins.
Jim Fleming
2002:[IPv4]:000X:03DB:...IPv8 is closer than you think...
http://ipv8.dyndns.tv
http://ipv8.yi.org
http://ipv8.dyns.cx
http://ipv8.no-ip.com
http://ipv8.no-ip.org
http://ipv8.no-ip.biz
http://ipv8.no-ip.info
http://www.ntia.doc.gov/ntiahome/domainname/130dftmail/unir.txt