Re: [v6ops] I-D Action: draft-ietf-v6ops-ula-usage-recommendations-02.txt

[Brian E Carpenter <[e-mail]>]

On 18/02/2014 08:45, joel jaeggli wrote:
> On 2/17/14, 11:24 AM, Brian E Carpenter wrote:
>> On 18/02/2014 06:11, joel jaeggli wrote:
>>> On 2/17/14, 8:38 AM, Ted Lemon wrote:
>>>> On Feb 17, 2014, at 9:35 AM, Owen DeLong <[e-mail]> wrote:
>>>>> Cases that propose NPT, NAT, or other translation mechanisms are
>>>>> examples where it is detrimental.
>>>> This is clearly true, but we can't do much about it other than to
>>>> encourage people to avoid this, and help them to avoid it by giving
>>>> them alternatives.
>>>>
>>>>> Cases where it ends up getting routed amongst "cooperating" parties
>>>>> are likely to eventually lead to detrimental usage because the
>>>>> natural trend is for them to become increasingly routed across more
>>>>> and more ASNs and eventually to become a form of de facto
>>>>> registered address space not administered by any structured or
>>>>> reliable registry and without any form of community based policy
>>>>> process (such as the current RIRs).
>>>> This is a fun doomsday scenario, but we don't have this with RFC 1918
>>>> addresses today; why would we have it with ULAs tomorrow?
>>> we route rfc 1918 between cooperating parties today. as noted it
>>> requires coordination and is painful.
>> Yes, but it's painful because of ambiguity. With ULAs all you have to
>> coordinate is which prefixes are announced between the cooperating
>> parties. No coordination is needed for prefix assignment, which
>> is the real hassle with RFC 1918.
> 
> you're assuming that humans allocate prefixes via the procedures
> described in the rfc 

People who don't will be punished automatically.

> and that utlization of a sparse address space is
> sufficently difuse the you don't have to worry about collisions or that
> more specfic routes will win out when someone has the bad taste to use
> fc00::/7 or 8 in a routing policy contribution to aggregate and so on.

People who do will be punished automatically.

> I kinda have my doubts. operational considerations around internal
> aggregation and the human need to group things together in sets as well
> as semantic uses (which I continue to assert are a bad idea) are going
> to cause these things to be used inside organizations on boundries other
> than /48.
> 
> My basic assertion is that if ula is widely and obiquitously  employed,
> owen's scenario is unlikely to come to pass because the results are
> messy and risky. If it isn't widely employed then who cares?

Exactly right.

    Brian