----------
From: Graham Klyne
Sent: Monday, March 29, 1999 8:25 AM
To: ifx@pwg.org
Subject: Online and connected?
My take on some recent exchanges...
One of the characteristics of traditional GSTN fax is that it's "circuit
based": signals travel on what physically appears to be a wire =
connection
from the sender to the recipient. If either machine is physically
disconnected then there's no communication, and nobody is surprised by =
that.
One of the characteristics of Internet access for a significant body of
users is that they are only intermittently-connected. Their equipment =
or
phone line is tied up for as long as they remain accessible via the
Internet (or, for many of us outside US, a phone bill meter is counting
away), so there is a natural tendency to hang up when the link is not
actively being used. The Internet equivalent of the physical circuit, =
the
TCP "virtual circuit", is effectively unplugged whenever the phone line =
is
hung up.
So far, so obvious?
Good. But note that the above has nothing to do with clients, servers,
dial-up protocols, or indeed any specific Internet protocol. It's just =
in
the nature of the way many people use the Internet today.
One of the attractions of e-mail is that it overcomes this =
intermittently
connected restriction, and delivers close to the best communication that =
is
possible under those circumstances. But this style of communication is =
not
compatible with the normal expectations of GSTN fax users.
This group (QUALDOCS) is attempting to address more directly the
expectations of GSTN fax users in an Internet communication environment.
If the plug is pulled (i.e. no active connection to the Internet) then, =
no
surprise, there's no communication.
There is one aspect of the current QUALDOCS goals that is fundamentally
inconsistent with intermittent connection to the Internet: timely
delivery. So, we are faced with a choice: do we ignore intermittently
connected users, or do we relax the goals in some way to allow them to
participate?
I take the above to be manifest. What follows is my opinion...
The only kind of mechanism that can degrade gracefully from =
fully-connected
to intermittently-connected cases is one that performs (or can perform)
message store-and-forward.
A corollary of this is that if IPP is to used then it will not, of =
itself,
address the needs of intermittently connected users. (This doesn't mean
that they cannot be addressed by additional future developments.) So =
maybe
the choice is really: do we defer consideration of intermittently
connected users as a later problem, or do we relax the goals in some way =
to
include them in the current work?
#g
------------
Graham Klyne
(GK@ACM.ORG)