Re: Operations requirement: no hop-by-hop

At 2:33 PM -0800 3/10/04, Dave Crocker wrote:
>Paul,
>
>PHI> The mail-ng transport protocol should be designed to only know about
>PHI> a single hop, from sender to recipient. The sender may have multiple
>PHI> hops to a designated transport agent, and the receiving transport
>PHI> agent may have multiple hops to a mailstore, but the transport
>PHI> protocol should not make any accommodation for these additional hops
>PHI> other than to allow them to add metadata to the transit stream. These
>PHI> requirements are to make the protocol more sensible and to limit the
>PHI> responsibility of operators to just what they control.
>
>How is this different from the current SMTP model?

In the current SMTP model, I believe we have been restricted from 
having SMTP options that "speak for" SMTP servers further in the 
chain. For example, we were prevented from making a "NO UCE" banner 
because there was no way for the server making that claim to know 
what SMTP servers behind it would have as policy. Different but just 
as complex logic was used in 8BITMIME.

I am proposing that we think of transport as just between two 
transport agents. What the receiving agent does with the message is 
not part of the protocol. Said another way, the transport protocol 
should not have rules that tell receiving agents what kind of policy 
decisions they need to transmit further down the line.

--Paul Hoffman, Director
--Internet Mail Consortium