- From: Willy Tarreau <w@1wt.eu>
- Date: Tue, 21 Oct 2014 11:44:25 +0200
- To: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Cc: Roberto Peon <grmocg@gmail.com>, Mark Nottingham <mnot@mnot.net>, RUELLAN Herve <Herve.Ruellan@crf.canon.fr>, Amos Jeffries <squid3@treenet.co.nz>, "ietf-http-wg@w3.org" <ietf-http-wg@w3.org>
Hi Poul-Henning, On Tue, Oct 21, 2014 at 09:32:55AM +0000, Poul-Henning Kamp wrote: > -------- > In message <20141021092505.GA30397@1wt.eu>, Willy Tarreau writes: > > >I would guess they should appear in the order above, though that's not > >obvious to me. And I'm still sad at the idea of leaving many encoding > >values unused (eg: static header values above 16). Thus, we'll typically > >have 48 possible values out of 256 for the first byte that will never be > >emitted just for the indexed headers alone, that's a 20% waste, > > If you are that worried about wasted compression opportunities, you > should spend your time to get timestamps compressed to integers since > that will save more bytes than you can ever do by tweaking the current > HPACK in any way. I know, and you remember, that was one of the basic points of our proposal 2 years ago. I think we'll hardly propose this here since it changes the ability to pass certain invalid values. However, I think we could do it by writing a proposal for a few header names and their binary representation that could be used as an alternative to Date/IMS/LM. A settings frame could indicate whether each side supports this extension and would only emit these ones instead of the regular ones. What do you think ? Willy
Received on Tuesday, 21 October 2014 09:46:03 UTC