Re: Operations requirement: no hop-by-hop

On Mon, Feb 16, 2004 at 08:32:51AM -0800, Paul Hoffman / IMC wrote:
> Please explain how you requirements above requires your MTA to tell 
> the sending MTA (or the sender, if the eventual protocol has no 
> initiating MTA) what the policies are ahead of time. Why can that not 
> be done at the time of message delivery, after the message's metadata 
> has been passed?

Because maybe the metadata to be passed depends on the policy.

Simple example:
I want to accept messages only if they are smaller than 10MB. This could
either be done
a) if we put the message size always in the meta data, or it could be done by
b) allowing the recipient to list SIZE as one of the meta data points he is
   interested in or
c) by specifying the limit in the protocol data send from the recipient to the
   sender.

A more complex example:
I want to accept messages only if the sender can prove that he is to be
trusted by using the Foobar signature scheme. Again I need to tell him
this before he will even try to send me a Foobar signature.

For the message size example it obviously wouldn't be a problem to
always transmit message size, because more likely than not most
recipients will limit this size and want to know it, and, its easy and
cheap to transmit this, only a few bytes more. But maybe the Foobar
scheme needs expensive calculation, so the sender will not just send
it in case it is needed, but only when the recipients specifically asks
for it.

So if we want some kind of pluggable policy structure that can be fit
to every recipients (and senders) need, we need some kind of negotiation
phase before the first message meta-data is even passed. This is
similiar to content- and language-negotiation in HTTP and to the cipher
negotiation in SSL. I am not sure whether we *really* want and need this,
but I can certainly imagine some scenarios where it might be helpful.

Jochen
-- 
Jochen Topf  [email protected]  http://www.remote.org/jochen/  +49-721-388298