- From: Graham Klyne <GK@acm.org>
- Date: Mon, 07 Jul 1997 12:41:05 +0100
- To: Koen Holtman <koen@win.tue.nl>
- Cc: hardie@thornhill.arc.nasa.gov, http-wg@cuckoo.hpl.hp.com
At 08:15 PM 7/4/97 +0200, Koen Holtman wrote: >If I understood your latest set of comments to the TCN draft right (I >am still working on writing a complete reply to these, and will send >it later), you seem to prefer a negotiation protocol which has >symmetric metadata, i.e. the metadata at the sender side has the same >format as the metadata at the recipient side. Yes. Mainly because I believe it will ultimately yield a simpler solution >TCN is very asymmetric with respec to metadata: you have variant lists >at the sender side and feature sets at the recipient side. However, >there are good reasons for these formats being asymmetric. They are >not asymmetric because of insufficient separation from HTTP. > >In TCN, negotiation is not really between two machines, it is between >a human being, a content author who writes a variant list, and a >simple machine (web browser). Intelligence-wise, this relation is >asymmetric (at least with the current state of browser technology). >It should therefore not be surprising that both parties have different >kinds of information to contribute to the choice process, and this is >reflected in the asymmetry of the metadata formats. This suggests that the assymmetry *is* caused by the relationship to HTTP. Not specifically the HTTP protocol, but the browsing application which HTTP supports. >In other types of negotiation, where two machines are involved, a >symmetric metadata format may be a nice win, but not so in TCN. In >TCN, it would just add to the complexity of defining the choice >process. I'm not convinced. But I can see I must produce some more concrete evidence for my view -- I'll think about it (it may take a while). (See also later comments.) >One could conceivably create a very general negotiation framework in >which both human-machine and machine-machine negotiation can be done >nicely. However, I do not have very high hopes that such a framework >could be created soon, or that it would be simple. > >I would therefore not be in favour of a `universal negotiation >framework' programme which would try to be applicable to both >author-machine negotiation (as in web content negotiation, or >selection of a part from multipart e-mail, or conditional HTML) and >machine-machine negotiation (as in fax or telnet option negotiation). > >On a related note, transport level simplicity considerations can be a >very dangerous guide when designing a negotiation mechanism. Case in >point: though the WILL/WONT/DO/DONT transport primitives in telnet >option negotiation are simple and symmetric, actually using them to >reach a decision turns out to be extremely difficult. A separate RFC >(RFC1143) was needed to fix the mistakes in the usage guidelines of >the original specification. RFC1143 is very instructive reading if >you are considering negotiation between two machines. It may be that we are talking very slightly at cross-purposes here. I would aim for a symmetric negotiation metadata format, but *not necessarily* symmetric transport protocol features to conduct the negotiation. In the case of HTTP and your comments about complexity of the choice process, I think the appropriate simplifying constraints could be enforced by the way in which the transport handles the negotiation (in effect, implementing a subset of the generic negotiation framework sufficient to the purpose at hand). Question: if I could offer (a) a generic framework and metadata format, and (b) indicate how these might be subsetted to current HTTP/TCN capabilities, would you consider that this would address your concerns? (NOTE: I am NOT proposing to re-design TCN or any part of HTTP!) GK. --- ------------ Graham Klyne GK@ACM.ORG
Received on Monday, 7 July 1997 04:45:35 UTC