Re: willchan's thoughts on continuations, jumbo frames, etc after *only skimming* the threads

On Fri, Jul 11, 2014 at 5:32 AM, Patrick McManus <pmcmanus@mozilla.com>
wrote:

>
> On Fri, Jul 11, 2014 at 2:47 AM, Rob Trace <Rob.Trace@microsoft.com>
> wrote:
>
>>  CONTINUATION is simpler than coding for some of the failure cases for
>> existing applications.
>
>
> I agree that unbounded headers are a compat issue - 16 bits is going to
> break some tail somewhere. So we need a continuation like mechanism.
>

Not quite:  You either "need a continuations-like mechanism", or you
increase the size to something sufficient.  If a 16MB header is still to
small, bump it even higher.  It seems self-evident to me that simply
increasing the size is the simplest approach.  It's easy to explain and
easy to implement.

Mike





>
> My first choice is to continue with h2-13 as is.
>
> My second choice would keep continuations but address some of the other
> issues
>  1] make frame sizes other than 14 bits configurable by an extension
>  2] get rid of the reference set based on the data, and require colon
> headers to be sent first and not in continuation
>  3] make hpack inside continuations stateless and allow them to
> multiplexed with other frames
>
> That does address the header hol blocker issues that current exist (though
> I don't think it is much of a practical problem which is why I am fine with
> proceeding without a change).
>
>
>
>

Received on Friday, 11 July 2014 14:51:14 UTC