Re: [BEHAVE] TCP port overloading, preservation and CGNs

[Simon Perreault <[e-mail]>]

Le 2013-06-25 19:20, ivan c a écrit :
>> Some NATs don't implement EDM and thus cannot switch to it.
>>
>
> By definition, all NATs are "EDM".

Please explain.

> "Endpoint-dependent Mapping" really means "Address and port-dependent
> Mapping". This is the most general case of NAT, where no requirement is
> made on the type of mapping. For instance, "EIM" is a particular case of
> this general case.

I don't understand how this could be true.

> Maybe we should use the term APDM instead of EDM for the sake of clarity,
> but the best is probably to avoid using the "EDM" acronym altogether if it
> leads to confusion.

EDM is defined in RFC 6887. Let's keep using already-defined terminology 
if it fits.

> Instead, we can use the term "no particular
> requirement
> on the mapping scheme" or something similar. This way, people new to the
> discussion can understand the point immediately and we also avoid the use
> of new obscure acronyms.

EIM and EDM are not requirements. They are qualifiers of NAT behaviour.

> Again, the goal here should be to suggest options for NAT to behave in p2p
> friendly ways. Some useful optional features for CGNs, like port
> overloading, do have minor caveats that should be detailed in the
> document,
> and they also don't apply to "3-tuple" NATs as you mentioned.
> This is the best way in my opinion to write a Best Current Practices
> document.

I would simply like to first understand technically what you're proposing.

Simon