Re: Large Frames, Continuations, Flow Control, and changing HPACK

> Jeff is proposing that we rearchitect the protocol.  Removing the
> reference set is a relatively inconsequential part of these changes.
> Being able to split header blocks changes the way that the protocol is
> structured; it changes the HTTP mapping.

"header blocks" and "header block fragments" as defined are a direct
consequence of the reference set. Without the reference set, these
things are meaningless.

> Greg has only proposed that we restructure the "jumbo" frame that we
> use for header blocks.
> From that perspective, Greg's proposal is far more a minor tweak.

True. It is minor, but it skirts around the underlying issue of HOL blocking.

> If the proposed change was to remove the reference set and maybe clean
> up some of the consequences of that, like the '\0' hack, then that
> would be a relatively small change.  It might also enable alternative
> solutions to the issues we're discussing (549, 550 and 551).  But if
> we're going to consider a major change like this, it needs to be
> better justified.

The question we pondered at the NYC meeting was "do we need to address
HOL blocking of headers frames."

At the time we decided that we didn't want to change the protocol to
address it. Since then there has been a lot of discussion on the list
on restrictions on HEADERS / CONTINUATION etc. because of that. This
is an attempt to address it directly.

Received on Tuesday, 8 July 2014 19:52:28 UTC