Feature negotiation requirements

Jim Gettys writes:
 > I've been watching the situation on content negotiation now for 18 months.
 > 
 > I believe the fundamental disconnect this working group has on the
 > topic is that there is no agreement on what the actual requirements
 > are;

Requirements often must evolve as the problem is further understood by
way of working through possible solutions.  So people should not
necessarily feel guilty about not having come up with the definitive
requirements up front.  But it does seem time to address the
requirements for content negotiation at a deeper level of specificity.

...
 > For example, I have a (personal) belief that only the client can
 > have enough information to make an informed choice, and that having
 > a proxy try to short circuit the first round trip will ultimately be futile.
...

 > But take this as just one opinion:  others feel that avoiding a round
 > trip to the first proxy is benficial (I do to; I just don't think you can
 > make it work well enough to be useful, and that the cost is worth paying).  

Perhaps one compromise requirement that everyone can agree on is that
round trips should be avoided, but not necessarily eliminated, without
increasing the size of requests or responses "too much".  This is
rather general, but I believe still useful as a guideline.  A more
specific, but arbitrary, goal of 80% round-trip avoidance seems like a
nice number.:-) This requirement can be statically achieved by doing
statistical studies of the most likely desired capabilities of servers
and clients.  But these capabilities will evolve over time, so the
content negotiation mechanism, or the way in which it is used, must
evolve in parallel so as to avoid round trips.

One of the problems we are bound to have is that the nature of
negotiation makes it is difficult for there to be very many absolutes.
The reason we need negotiation in the first place is that we have a
disconnect between two (or more) parties who wish to communicate about
something.  The process of negotiation must always be balanced against
the efficiency of the request process and the precision of the
results.

--
Daniel LaLiberte (liberte@ncsa.uiuc.edu)
National Center for Supercomputing Applications
http://union.ncsa.uiuc.edu/~liberte/

Received on Tuesday, 10 June 1997 11:11:52 UTC