Minutes of HTTP Working Group, 33rd IETF Meeting, in Stockholm

Reported by Jim Gettys from notes taken by Henrik Nielsen.

HTTP 1.0

We need to come up with a final draft before we can finish all discussion. Draft will be available August 1, for anticipated last call later in the month.

Access Authentication - MD5 digest

There is real no objections to the current state of this proposal. HTTP does not provide possibility for having MIME headers after the HTTP object.

There are multiple implementations:

NCSA server and client
Spyglass
Dave Kristol's server

HTTP Session Extension

Ted Hardie, NASA, led this discussion.

This proposal would avoid TCP latency, overhead, and slow start performance problems.
Ted described a proposal from Alex Hopmann, who was not present
Henrik noted that Request-ID header makes the proposal more flexible as the server can send them back out of order
There was general talk about sessions with a server.
Jeff Mogul of DECWRL has made an extensive study and simulation of persistent connection HTTP. The results of this work can be found at http://www.research.digital.com/wrl/publications/abstracts/95.4.html
Is it a good idea to save headers while a connection is kept alive?
- Eric Sink: No big advantage - 10% (from implementation)
- Larry Masinter: for almost all headers, it's a win; the only issue is those headers for which there is no way to say 'the default' by giving a header explicitly.
- authentication may be the biggest performance win.
A number of implementations were mentioned; performance is unclear, and most likely to be seen over long haul and dial up lines, rather than in a local network, where most naive tests are performed.
The general consensus is that persistent connections are a good idea. There are concerns about upward compatibility and interoperability with 1.0; this may or may not require 1.1; it was suggested that operation under 1.0 might be written up as an experimental protocol.
An open question is the timeouts for the tcp connection; there is some data from Jeff Mogul's simulation.

MIME multi-part

MIME multi-part was not discussed.

Session-ID, Request-ID, cookies

No one wanted to talk about it at this meeting.

Problem with HTTP PUT and POST

Henrik Nielsen described a problem with HTTP PUT and POST that has recently been uncovered, and solicited feedback.

HTTP/1.1

A HTTP/1.1 draft will be available in mid-August.

HTTP/NG

HTTP/NG: Andy Norman, Ange@hplb.hpl.hp.com. Stefek Zaba has an experimental implementation of what they call HTTP/NG, and has been taking to Simon Spero. Simon was not at the meeting, so there was little discussion of NG.

Larry Masinter pointed out that people are trying to do transactions with HTTP when HTTP does not have a transaction mechanism (e.g. when you try to abort an operation in the middle, you have no way to know whether or not it completed.)

Many RPC implementations do what people are trying to do with HTTP-NG: keep connections open, let them time out, handle more complex operations, interleave multiple calls and results on the same connection.

This begs the question: Who has implemented a non-blocking (streaming) RPC system that can be used if we are to avoid rolling our own? Does it have the needed facilities?

Feedback from John Klensin, and Harald Alvestrand, Area Directors for Applications

John Klensin expressed great displeasure with the current state of the working group. Some issues he raised, but not necessarily an exhaustive list include:

Working group chairs that do not warn the area director before an IETF meeting that they cannot attend are asking to be shot. John promised to convey his displeasure directly to the chairs.
How will we make progress?
We have a collective problem in the working group. We should stick to the milestones.
Without NG as a milestone for this group, 1.1 will likely end up out of control. Without Simon Spero present, and with his RSI problems, John is very concerned about NG. Jim Gettys volunteered to edit HTTP/NG, if Simon is unable to deal with it due to his problems. When will it become a proposed standard?

Harald Alvestrand noted that there is no reason to wait for an IETF meeting in to send a document to the IESG for standardization.

Proposed New Milestones

AUG 95 send HTTP/1.0 of to IESG as proposed standard.
OCT 95 Session as experimental extension.
APR 96 HTTP/1.1 as proposed standard
DEC 96 HTTP/NG as proposed standard. Jim Gettys volunteered to help Simon with writing.