On 3/17/2004 5:16 PM, Lyndon Nerenberg wrote:
Submission is by definition an authenticated service that authenticates
to a specific > set of servers, and therefore must be hand-configured
into the client. (And if there are several servers the client can
submit to, a CNAME will solve that problem.) This only needs to be done
once, so I don't understand why it's even an issue.
Simplifying the one-time configuration, for one thing. Instead of asking
mom to answer a bunch of questions like server, port number, TLS, etc.,
just ask for her email address and password, and see if the config can
figure out the rest.
Recovery from failure is another. If the static server goes bye-bye for
whatever reason (potentially including roaming), the client should be able
to detect the mechanisms associated with their home network.
On a side point, the mechanisms in the draft describe RETRIEVAL services
as well as submission services. While we can talk about submission
mechanisms that might be allowable for certain situations, retrieval is
almost universally locked to specific servers, and is not roamable.
Eric A. Hall http://www.ehsco.com/
Internet Core Protocols http://www.oreilly.com/catalog/coreprot/