Tony (et al),
I suspect there is a key point of design (architecture) being missed,
here. A note from Ned Freed asserted the "operational" implication of
this architectural point, but just in case the underlying framework is
A sender needs to specify an "address"; the transport system needs to
route it; and the receiver needs to be ready with server arms open and
waiting. This model can be made more complicated by having the routing
be a multi-phase thing, so that the decision process isn't specified
globally. But some set of decisions have to figure out how to get from
sender to receiver. (Stupidly basic stuff, I know, but bear with me.)
In Internet email, addresses are of the formal
The semantics of the global part are truly global. ANYONE can parse
and comprehend them. The semantics of the local part are truly local.
NOONE but the entity referenced in the global part is allowed to
believe it has the slightest clue as to the meaning of the local-part
What this means is that:
1. Placement of the telephone number in the global string means that
the global "routing" system, in the form of DNS registration, lets
different servers plug in in different places, without any particular
coordination among themselves. Further, it means that the sender needs
no special knowledge. They specify the phone number, as a classic
"address" (which says where a thing is but not how to get to it) and
let DNS resolution and IP routing figure out the rest.
2. Placement of the telephone number in the local string means either
that the global part needs to have its own special semantics or that
the global part is used only for partial routing and that the local
part then must be further processed, to send the message on, to the
real final destination. In the attmail example, note that the sender
must know to specify attmail. The fact that attmail has multiple fax
servers is special knowledge to attmail and is handled as part of the
"further" processing I cited. Once we start having these servers
provided by a very large number of independent organizations, then I'm
not sure how we do the addressing. Remember that the global-part is
the only place we should put "global" addressing information and
remember that NOONE is allowed to interpret the local-part, other than
the global-part referent.
So, alternative 1 results in some astonishingly clean behaviors:
1. Standard format for ALL fax. The only thing that varies is the
2. No redundant addressing information in the string.
3. Scalable at the whim of the servers. Users are oblivious.
4. Variable scope for servers. If I want a server just for my home, I
can register it and don't have to get anyone's permission except
for regular DNS registration. The same holds for country-level
service provision, for the really ambitious servers.