[v6ops] (re)numbering [ULA draft revision #2 Regarding isolated networks]

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

Hi Mark,

On 28/05/2014 10:37, Mark ZZZ Smith wrote:
> Hi Brian,
> 
> 
> ----- Original Message -----
>> From: Brian E Carpenter <[e-mail]>
>> To: Mark ZZZ Smith <[e-mail]>
>> Cc: Liubing (Leo) <[e-mail]>; v6ops WG <[e-mail]>; "[e-mail]" <[e-mail]>
>> Sent: Tuesday, 27 May 2014 8:40 AM
>> Subject: Re: [v6ops] ULA draft revision #2 Regarding isolated networks
>>
>> Mark,
>>
>>
>> On 27/05/2014 09:57, Mark ZZZ Smith wrote:
>> ...
>>>>  - "Temporarily isolated" or "Forever isolated". In 
>> general, 
>>>>  ULAs fit both cases. Whatever it is temporarily or forever, when 
>> administrators 
>>>>  need some prefixes to be on-demand and free to use, ULAs are good 
>> choice. 
>>>>  However, for the temporarily isolated cases, the administrator needs to 
>> consider 
>>>>  once it gets to connected, the hosts might need to be renumbered; or 
>> NAT might 
>>>>  be involved if renumbering is not acceptable. If renumbering or NAT for 
>> some 
>>>>  reason is considered as heavy burden, then the administrators need to 
>> carefully 
>>>>  consider the adoption of ULAs.
>>>>   
>>>  This paragraph seems to show a fundamental misunderstanding of IPv6's 
>> multi-addressing capabilities. IPv6 supports multiple concurrent addresses (from 
>> different prefixes), and can learn new ones or deprecate old ones over time. 
>> Attachment to a new network doesn't require renumbering, it requires 
>> propagating new prefixes for the hosts to use in addition to their existing 
>> ones. Primarily RFC6724 address selection will help the hosts choose the right 
>> addresses to use as source and destinations when they have multiple addresses.
>>
>> I didn't read it that way. Of course an IPv6 network runs well with
>> multiple prefixes, which is why overlapped renumbering is possible.
>>  
> 
> I think there are a number of reasons I read it that way.
> 
> Firstly, to me, the word 'renumber' reads as 'replace the numbers', so renumbering a network involves just that - removing the old numbers and adding new ones. The suggestion of using NAT has been in IPv4 the alternative to 'replacing the numbers', so suggesting that for IPv6 seemed to further support the misunderstanding of IPv6's ability to support multiple concurrent addresses/prefixes/spaces.

Whereas in RFC 4192 it means exactly the opposite: adding new numbers and
then removing the old ones. That really is a change in thinking for IPv6
and of course you are right - multi-addressing is not understood in the
industry.

> The other reason is that I've actually seen a residential CPE try to swap between global addresses and ULAs. I deal with a number of residential IPv6 CPE vendors in around 2009, and one of them had attempted to support ULAs, but had not done a good job of it. When the global prefix went away because the WAN link failed, the CPE would try to both flush the global prefix by setting a 0 valid liftime (or similar, I can't quite recall) and replace it with a ULA in the RAs on the LAN side. When the WAN link came back, it tried to do the opposite. When I provided feedback to this vendor on this issue and others related to their IPv6 implementation, they just ignored it, unlike the other CPE vendors I was dealing with.

That's horrible. It doesn't conform to RFC 7084 either, which clearly
states that "prefix(es) (and ULA prefix if configured...)" must
be advertised.

    Brian

> Regards,
> Mark.
> 
> 
> 
> 
> 
>> As Fred Baker once pointed out, the real problem is therefore
>> *numbering* a network (i.e. adding a new prefix, regardless of
>> whether there are zero or more existing prefixes). The text needs
>> to be clear about that, for sure.
>>
>>     Brian
>>
> .
>