Re: Reference set in HPACK from Martin Thomson on 2014-07-02 (ietf-http-wg@w3.org from July to September 2014)

From: Martin Thomson <martin.thomson@gmail.com>
Date: Wed, 2 Jul 2014 11:27:16 -0700
To: Mark Nottingham <mnot@mnot.net>
Cc: Kazu Yamamoto (山本和彦) <kazu@iij.ad.jp>, HTTP Working Group <ietf-http-wg@w3.org>
Message-ID: <CABkgnnXWZzk-EG2EcQs1Bve0pYyAqmk91zqrJheY4LJUB+rP4w@mail.gmail.com>

On 1 July 2014 22:44, Mark Nottingham <mnot@mnot.net> wrote:
> As I understand it, one of the purposes of the reference set is to accommodate cases where intermediaries multiplex requests from different clients -- thus having different cookies, UAs, etc. -- into one upstream connection.


It's kinda the opposite actually.  The reference set clear is for that case.

Reference set is really good for server-to-server cases where the
request header fields are basically constant inter-request.  In those
cases, the marginal cost depends on how many headers are present, but
the difference is approximately 1 byte per header.

Received on Wednesday, 2 July 2014 18:27:44 UTC