Re: [Autoconf] Working Group Last Call for draft-ietf-autoconf-addr-model-01

[Mark Townsley <[e-mail]>]

cc'ing Dave Thaler directly, I need his advice here. Please see inline...

On 1/20/10 2:51 PM, Thomas Narten wrote:
> Mark,
>
>
>    
>>> I don't see a lot of difference between an IPv6 "on-link" prefix and
>>> an IPv4 subnet prefix. It doesn't make sense to me that IPv6 and IPv4
>>> are treated differently.
>>>        
>    
>> I think I was the one who actively supported treating IPv6 and IPv4
>> differently in the document. IPv6 has a ULA range, IPv4 doesn't. IPv4
>> has a Private range, IPv6 doesn't.
>>      
> This isn't a real difference. Net 10 is "private" in IPv4 and spans
> multiple links/subnets. That is no different than a ULA in IPv6.
>
> Yes, there are some important differences between ULAs and net 10
> addresses (namely global uniqueness), but that doesn't change how
> prefixes out of those networks are assigned to individual links, and
> whether they are to be treated as being on-link (or not), which is
> what is at issue.
>
>    
>> IPv4 global space is constrained, IPv6 isn't. IPv4 stacks remove
>> IPv4 link-local addresses from an interface when a global or private
>> is assigned, IPv6 stacks do not...  etc... etc... So, while there
>> certainly are commonalities in the models, there are enough
>> differences that describing each model separately seems to reduce
>> the chance of misunderstanding the differences that do exist between
>> one version vs. the other.
>>      
> This doesn't really answer my core question. There is no real
> difference between a /32 in IPv4 and a /128 in IPv6. What is the
> justification for saying /32 should be used in IPv4 but no on-link
> prefix is used in IPv6?
>    
Indeed, in the previous draft we had this text:

>    o  (An IPv6) subnet prefix configured on this interface should be 
> of length
>       /128.
> and
>
>    o  Any (IPv4) subnet prefix configured on this interface should be 
> of length
>       /32.

However, Dave convinced us at the last IETF that the better way in IPv6 
to describe what we really want to say is that on-link subnet prefixes 
should simply not be configured at all. Practically speaking, I don't 
think we have this same luxury in IPv4.

- Mark
> BTW, I'd be fine with having the IPv6 section say on-link with a
> /128. That is what I would have expected for consistency.
>
> I don't think doing this is a big deal, unless you have IPv4 and IPv6
> be different, with no good technical justification, which is how I see
> things at the moment.
>
> Thomas
>
>