On 12/1/11 20:28 , Pete Resnick wrote:
On 12/1/11 10:12 PM, Doug Barton wrote:
On 12/01/2011 19:47, Pete Resnick wrote:
The current draft says that the reason 1918 space can't be used is that
equipment that deals in 1918 address space is hosed if 1918 addresses
are used on their external interface.
Let's assume that's true for a second (I haven't seen any evidence of
that). We all know that if the /10 is allocated that people are going to
use it for 1918 space. So, back to square 1.
No, that's not true. Once this document claims that a particular block
of addresses will be used on both internal and external interfaces,
whether they're from a part of 1918 space that isn't used by the broken
equipment *or* from a new /10 (which obviously isn't used by the broken
equipment), any *new* use of this address space by *newly* broken
equipment is acceptable to the CGN people. The only thing the current
document worries about is deployed equipment that the CGN people can't
push back on. So either a new /10 or 1918 space not used by current
broken equipment addresses this problem.
Brian claimed that perhaps
172.16/12 space might not be used by that equipment. Robert claimed that
perhaps only 192.168 and 10.0.0.x addresses are used by that equipment.
So the question I posed was, "Does any of *that* equipment use 172.16/12
(or 10.x/16) space?" Nobody has said "yes".
And *I'm* still not claiming that the answer is "No." I simply don't
know. But I'm inclined to hear from anybody to indicate that there is
*any* evidence that the answer is "Yes". That would make me much more
comfortable in concluding that new specialized address space is the
better horn of this bull to throw ourselves on.
The lack of research on this point has been pointed out in the past, and
TMK has never been addressed.
Ron's point (that part of the problem is that people simply don't know)
is well-taken, but if there is not even anecdotal information that
172.16/12 or 10.x/16 is used by broken equipment, I'd like there to be
some research before we say that allocating a /10 is necessary.
it's simpler than that.
assuming that there existing a pool of devices for which it can be
* does not support a collision between internal and external
* has a collision between it's internal address ranges and
assigned external prefix in 10/8
It seems unlikely in the extreme that home cpe statifying both
conditions would also have a collision with an assignment out of 172.16/12.
I have never found the arguement that, that particular problem
intractable enough to benifit from and additional prefix to be
Ietf mailing list