DRAFT Minutes, HTTP working group

These are draft minutes of the HTTP working group meetings held this
week at IETF. Ted Hardie took minutes; I've edited a bit based on my
own notes and memory. If you were at the meeting, please review as
quickly as possible so we can send these in.

<<If you want to comment on the topics actually discussed at the
meeting, please use a new "subject:" in your message.>>

Minutes for the HTTP Working Group, 36th IETF, reported by Ted Hardie.

The minutes reflect the topics, discussion, and decisions made at the
working group meetings over the two days, although not in the exact
order they occurred.

Related work:

WTS had submitted its drafts for last call and was not meeting at this

The index BOF met Monday. It began with a report of the W3C sponsored
workshop on web indexing. This work is of interest to HTTP
development, but does not currently look like it will result in a
working group forming.

A URN BOF will meet Thursday.  This is important to HTTP-WG members
since the use of URNs might also have an impact on caching models
within HTTP.

There is some work starting on versioning on the web. There will be a
workshop in July on the topic, held at NaviSoft (San Mateo, CA).

Content negotiation:

There are three current drafts that deal with negotiation some form
and content negotiation in particular (transparent negotiation, UA
headers, and PEP).

The group is interested in the work Koen has done, but worried about
the timetable for filling in some of the important pieces--a volunteer
was requested to help Koen with completing his model and draft.  No
volunteer was available at the meeting, but interest remains if
someone can commit to working on it.

Andrew Mutz described the draft on UA headers. The reception was
mixed.  Some members of the group felt that it provided a good,
standard method for providing information about UA capabilities which
might not otherwise be available (particularly in the context of PDAs,
braille terminals, or voice output).  Others felt that this was the
thin edge of a wedge for a host of UA headers which did not belong in
a transport layer protocol.

A discussion of the use of a generic UA header revealed problems with
caching and limitations on how Vary headers could be attached; the use
of MIME envelopes to accomplish the same effect was also proposed.

Henrik reviewed the status of PEP. In his talk, he pointed out how PEP
could be used for the UA negotiation of the Mutz draft as an example.
He believes that the continuing use of RFC822 headers, while easy, is
not effective as it cannot handle order dependency nor the difference
between need and want.  He noted that many of the previous problems
with PEP (lack of a class hierarchy or q factors, etc) are being
addressed now within the W3C during test implementations.  A new draft
will be available with those changes by August 1, 1996.
It is important to describe how PEP will interact with caching,
especially with older caches; worries that this sort of mechanism will
radically increase the number of cachable variants are a work item.

Harald Alvestrand then pointed out that the group does not have a
unified model; we have a desire to create a language to describe what
the user wants and a language to describe what the server has and we
don't have a unified model for bringing those together; until that
model comes together--neither is going forward.

We will continue to work on all three drafts; the final decision will
be deferred until there are revised drafts and implementation
experience is available.  The question is not really whether or not to
support variability--variability is already being done based on
User-Agent and other factors, but we do need a model for handling it
long term.

Keith Moore then pointed at that we need to be careful about how far
down this path we go; some of what we want is outside of the box of
HTTP, and that we don't want to set things up so that we have to use
HTTP for everything or bootstrap everything through HTTP.  Look for a
coherent method of handling the near and middle-term problems.


We discussed the three drafts originally submitted by Philip
Hallam-Baker.  The group agreed that demographics were important,
because people sometimes deliberately defeated caching in order to get
good demographics, but that many demographic methods actual fell
outside the scope of the working group because they do not require
anything be done to the protocol.  Jeff Mogul and Paul Leach agreed to
put forward a draft by the end of July which would describe a simple
method for doing a very baseline user/hit count with an extension to
the cache-control directive.  It was also suggested that a
requirements document for demographics would be useful, as many had
different ideas of what was needed.  Among them were:

user counts
hit counts
session identifiers
client capability counts

John Klensin and Keith Moore then reminded the group that many
business models do not allow for the automatic sharing of data, and
that the notion that anyone can cache information should not be a
basic part of our model.  If do work on demographics and interactions
between servers and caches, we should do so with the understanding
that this work is limited to situations where a trust relationship
exists between the server and cache.

Previous work:

draft-ietf-http-digest-aa-04.txt (Digest Authentication) will go to
last call immediately. Most (but not all) of the implementors in the
room agreed that they would implement it.

draft-ietf-http-state-mgmt-02.txt (State Management) will need some
editorial changes before being sent for last call, and Dave Kristol
will be producing a new draft.  This draft will remove the word
Proposed from the title, and clarify certain other potential areas of
concern. This work will happen within 3-4 weeks. Marc Soloman is
working on some other issues with Dave on this.

Implementation Guide:

Martin Hamilton agreed to gather items from the current draft which
might be better described in an implementation guide; he will create a
web page of this, and post the URL.  Anyone interested in created an
RFC or Book on the subject is encouraged to do so.  The group also
agreed that the creation of this document would not mean removing all
explanatory material from the draft.


The group agreed that the current handling of comment text for
matching on Vary fields was conservative (since only LWS might be
ignored), but decided that it would be easier to begin with this model
and loosen it later, if necessary.

The issue of how to handle character sets was raised, in that the
current draft's handling does not match the IAB workshop's general
recommendation on character sets (though the workshop did include a
"grandfather" clause for existing protocols).  After a great deal of
discussion, it was agreed that a small group (Larry, Paul Leach,
Francois) to generate new wording which requires a charset label on
all text types, and which explicitly uses the "unknown" charset when it
is not known.

At the close of the discussion of 1.1, Jim Gettys was hailed with
thunderous applause for his work on getting 1.1 ready.


Work on HTTP-NG is ongoing, including work on a multiplexing protocol.
Jim Gettys made an appeal for data with which to determine what
problems need to be solved for HTTP-ng.  Without data we are likely to
work on the wrong problems. It was suggested that if someone wants
data, the right process is to write an RFC suggesting what kind of
data should be gathered and where it should be sent. Jim will write a
short document describing what data is needed and giving a limited
time period in which it will be accepted.

HTTP/1.2 issues:

There are only a few issues that people want to see resolved in HTTP
for which we do not currently have Internet-Drafts.  Paul Leach has
agreed to write a draft on sticky headers, compressed headers, and
context identifiers.

Michael Ottati asked whether there was interest in the group in
creating a standard method for handling atomic transactions (a commit
facility).  Those interested should contact him at
Ottati_Michael@tandem.com; he will also write up a draft describe the
basic facilities he's interested in, for comments by the group.

Revised Charter

The group needs to revise the charter. We identified the following

July 2: Revised HTTP/1.1 draft with
    charset labelling
    ;q= tag changes

Aug 1: Henrik:  PEP draft
Aug 1: Mutz:    revised User Agent draft
undated:  revised content negotiation draft
undated:  HTTP Implementation Guidelines draft

Aug 1: Leach: draft on sticky headers, short names for headers, and
	context identifiers

July 30: Mogul, Leach: draft on 'hit count' additions

Dec 96: all remaining documents to Last Call

Target: close working group by next IETF with work completed
	may not need meeting at December IETF.

Received on Wednesday, 26 June 1996 21:06:50 UTC