Re: #540: "jumbo" frames

On 27 Jun 2014, at 5:21 pm, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote:

> In message <7D8D6BB1-1527-4C79-9E41-BBBF6293B61E@mnot.net>, Mark Nottingham wri
> tes:
> 
>> I think this discussion has converged upon not making any changes to the =
>> HTTP/2 spec, but allowing experimentation to take place in a "jumbo" =
>> extension.
>> 
>> As such, I'm going to close the issue. If implementation and deployment =
>> experience in the next round leads us to think differently, we can =
>> revisit the question, of course.
> 
> I would like to propose that CONTINUATION be removed to an extension
> as well then.

Can someone please provide a proposal? Some features of the protocol aren't amenable to being shifted to extensions, and I very much suspect that CONTINUATION is one of them...


> Several people here have already said that they are not going to
> implement CONTINUATION and there is presently no way to negotiate
> if CONTINUATION will be available or not end-to-end.

There's a slight difference here -- people have said that they'll refuse requests with CONTINUATION, which can be seen as part of the choices they make when defining the protocol elements (and sizes thereof) that their implementations will accept.


> CONTINUATION seriously inflates and complicates the draft-RFC
> with its many mysterious and counter-intuitive rules (Such as
> "END" flags which suddenly don't mean "END" and so on.)
> 
> There is also no published implementation experience with CONTINUATION
> that shows it to be a particular good solution to anything in particular
> or for that matter at all.

I'm just finishing drafting the announcement of the next implementation draft and WGLC, the plan being to allow the spec to stabilise for a longer period of time, so as to allow both implementation and deployment. That should give better evidence one way or another.

Cheers,

--
Mark Nottingham   https://www.mnot.net/

Received on Friday, 27 June 2014 07:36:12 UTC