Roland,
On 2010-10-26 20:57, Roland Bless wrote:
> Hi Brian,
>
> Am 26.10.2010 00:30, schrieb Brian E Carpenter:
>> On 2010-10-26 04:59, Gert Doering wrote:
>>> This discussion has been rehashed a number of times now, and it's time
>>> that the "anti-NAT" crowd starts to accept that e2e is not a desirable
>>> property in some networks, and thus, this aspect of NAT doesn't do "harm".
>
>> The problem comes when one of the ends tries to participate in
>> a multi-party protocol. The state that a NAPT creates to permit
>> a two-party protocol to work isn't able to support a third party.
>
>> So, people whose model of connection to the Internet only involves
>> two-party client-server protocols can use the arguments Chris Engel
>> has expressed, but if they want multi-party protocols they have to
>> start using some kind of kludge. (I am including things like ICE
>> in the category "kludge".)
>
> But you also have to use a kludge in case your protocol carries
> IP addresses inside. While it's well known to better avoid that,
> some protocols cannot live without it, e.g., transport protocols.
> So new transport protocols or other applications will still have
> a hard time when it's time to deploy them. Hence, multi-party protocols
> are IMHO not the only problem.
Correct, although our architectural recommendation against carrying
IP addresses in upper layer protocols has existed for just about
as long as NATs have existed.
This is exactly why I'm working on draft-carpenter-referral-ps, which
is intended to include the two-party case as a subset.
Brian
_______________________________________________
nat66 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nat66
