Re: [nbs] NBS and TCP connection identification

[Rémi Després <[e-mail]>]

Christian,

The semantics of the IP address at the socket API can, and IMHO should, remain unchanged during the connection.
I see no need to invent synthetic names just for the application to be able to find again the initial address.
If the name of the connection initiator is not given received in the connection-accepting host, this one should not accept the connection if it imposes to have the initiator's name (one that can be checked in the DNS).

Regards,
RD 

Le 12 oct. 2010 à 18:58, Christian Vogt a écrit :

> 
> You wrote:
> 
>>> Needless to say, if the initiator does not provide a name -- e.g., because it does not have a name registered in the DNS --, the responder would generate a name for the initiator based on the initiator's IP address.  Such a "synthesized" name does not have to be verified; it already provides the same level of security that we have today.
>> 
>> It isn't clear to me that synthesized names are needed. 
>> NBS APIs could instead continue to provide, in their binary format, addresses found in session originating packets, and let application ignore them if they aren't interested.
> 
> Remi:
> 
> The reason why I think we do need synthesized names is because of their semantics:  They name the host that was located at the encoded IP address at the time of session initiation.  Unlike the semantics of an IP address, which can name the host that /currently/ is located at the address, a synthesized name could be stable across a session.

> 
> Does this make sense?
> 
> - Christian
>