Re: [dcp] Features

[Eddie Kohler <kohler@icir.org>]

Hi Sara, my take:

> One way to remove some negotiations would be to make it unnecessary to
> establish the value of features which are mandatory for a particular
> CCID through negotiation, let them be determined by the CCID instead.
> Such an example would be the Ack Vector for CCID2. When you have opted
> for CCID2, the Ack Vector could be included automatically.

You are right that several negotiations might be required, although they
could be largely overlapped, but I bet this won't happen very often. For
instance, the usual sequence for a server S choosing CCID 2 would look like
this:

      C->S: Ask CC 2
      S->C: Answer CC 2; Ask Use Ack Vector 1
      C->S: Answer Use Ack Vector 1

Note that Use Ack Vector isn't negotiated until the end. S asks the client
C to use the Ack Vector, so that request can piggyback on the Answer.

We negotiate the features independently from CCID to facilitate
extensibility. This way, a new CCID can be deployed without upgrading every
DCP implementation in the world; it'll request the features it needs.

Your solution would certainly help if long negotiations became a problem in
practice.

> Are you planning on enforcing a certain
> behavior
> so that one end-point has more influence?

We weren't planning on it. CCID negotiation is a negotiation among equal
parties.

> I have another issue which I would only like confirmed; at connection
> set-up is it enough to agree on the CCID to start sending data, the rest
> of the negotiations can be concluded while sending data?

You're right. To clarify, though: The draft says you cannot send data when
you don't have a valid CCID. An implementation, or a particular CCID, might
further hold off until some additional features are negotiated. That would
depend on the implementation and the features. For example, CCID 2 might
refuse to send data (or more than an initial window of data) until the
receiver agrees to Use Ack Vector.

Eddie

_______________________________________________
dcp mailing list
dcp@ietf.org
https://www1.ietf.org/mailman/listinfo/dcp