Re: [pcp] PCP Server Selection - Address Family Selection

["Prashanth Patil (praspati)" <[e-mail]>]

We've tried to simplify text around this in the revised draft. Hopefully,
it is more obvious now.

-Prashanth

On 4/28/14 6:10 PM, "Simon Perreault" <[e-mail]> wrote:

>Le 2014-04-27 20:52, Dan Wing a écrit :
>> 
>> On Apr 26, 2014, at 2:37 PM, Sebastian Kiesel <[e-mail]>
>>wrote:
>> 
>>> On Fri, Apr 25, 2014 at 06:42:53PM -0700, Dan Wing wrote:
>>>>
>>>> On Apr 25, 2014, at 7:57 AM, Simon Perreault
>>>><[e-mail]> wrote:
>>>>
>>>>> Le 2014-04-25 10:51, Sebastian Kiesel a écrit :
>>>>>> draft-ietf-pcp-server-selection-02  says in Section 3:
>>>>>>
>>>>>>  1.  If the PCP client can use both address families when
>>>>>>      communicating to a particular PCP server, the PCP client SHOULD
>>>>>>      select the source address of the PCP request to be of the same
>>>>>>IP
>>>>>>      address family as its requested PCP mapping (i.e., the address
>>>>>>      family of the Requested External IP Address).
>>>>>>
>>>>>>
>>>>>> What is the reason for saying it SHOULD use the same familiy as the
>>>>>> requested EXTERNAL IP address?  I think it would make more sense to
>>>>>>use
>>>>>> the same familiy as the INTERNAL address (if there is a difference
>>>>>>at
>>>>>> all).
>>>>>
>>>>> +1
>>>>>
>>>>> I stated this in my review.
>>>>
>>>> The intent of the wording is to avoid NAT64 (or NAT46) where NAT44 or
>>>> no NAT was necessary at all.
>>>
>>> reasonable goal, but does text about that belong here?
>>>
>>> this text is about PCP messages, and I think they should use the
>>> same address family as the actual user data flow on their side of
>>> the NAT.
>> 
>> If that avoids IP address family translation, sounds good to me.  Or is
>>there some other value or purpose?
>
>I understand what you want, but the text doesn't get you what you want.
>I want something different, but we can both get what we want with
>appropriate text.
>
>First, "sending over IPvX" doesn't imply that the internal address is
>IPvX, because of THIRD_PARTY. What you want is: "the client SHOULD pick
>an internal address of the same family as that of the mapping's external
>address."
>
>Second, what I want is to avoid THIRD_PARTY. So that's why you need to
>send the request using the chosen internal address as source for the PCP
>request. (This is actually stricter than just using the same address
>family, in case you have multiple addresses of that address family.)
>
>Summarizing:
>
>"If a PCP client has access to internal addresses of multiple families,
>then it SHOULD choose the mapping's internal address such that it is of
>the same family as that of the mapping's external address. This is done
>to avoid address family translation when possible.
>
>The PCP client SHOULD use the mapping's internal address as source for
>the PCP request. This is done to avoid using the THIRD_PARTY option when
>possible."
>
>Simon
>-- 
>DTN made easy, lean, and smart --> http://postellation.viagenie.ca
>NAT64/DNS64 open-source        --> http://ecdysis.viagenie.ca
>STUN/TURN server               --> http://numb.viagenie.ca
>
>_______________________________________________
>pcp mailing list
>[e-mail]
>https://www.ietf.org/mailman/listinfo/pcp