Resent-Date: Thu, 03 Nov 2016 10:32:53 +0000
Resent-Message-Id: <E1c2FKH-00034q-SJ@frink.w3.org>
MIME-Version: 1.0
In-Reply-To: <568945A1-75B6-4E9D-9021-38479FC55580@cloudflare.com>
References: <147793576451.32369.14134057573457350871.idtracker@ietfa.amsl.com>
 <3669167D-26AC-4B78-8175-99B0028B6891@cloudflare.com>
 <CABkgnnXqHP6RNpHBcFStO5TWz8Sq6Uqs7KMWFof88RjxhoW-Qg@mail.gmail.com>
 <06396a0d-a0c1-19fc-85d5-6ddfb9bcf39f@gmx.de>
 <CABkgnnWFds=rYHc-ufCynXg701ekQ6MJTrbXXZrV0ozRod6HzA@mail.gmail.com>
 <D8E74F06-A6CC-4EA9-9D7C-EFD043F72624@cloudflare.com>
 <CANatvzzZOvPWrdQqNfV4VSiZ4cb2zt36f1-mKTrxTS8kW6eSuw@mail.gmail.com>
 <CACweHND+E7D0oKR+_2sKVOqrAwx_hQW9Z=MAmDFGfbqEzR4xGQ@mail.gmail.com>
 <568945A1-75B6-4E9D-9021-38479FC55580@cloudflare.com>
From: Jyrki Alakuijala <jyrki@google.com>
Date: Thu, 3 Nov 2016 11:32:14 +0100
Message-ID: <CAPapA7TfTMdOA4XB=9b6959CUxqfZFe=A8nVnQB3xd3pG8HBRg@mail.gmail.com>
To: Vlad Krasnov <vlad@cloudflare.com>
Cc: Matthew Kerwin <matthew@kerwin.net.au>,
 Martin Thomson <martin.thomson@gmail.com>, 
 HTTP Working Group <ietf-http-wg@w3.org>
Content-Type: multipart/alternative; boundary=001a113eec1a3f8ac6054063126b
Received-SPF: pass client-ip=209.85.218.41; envelope-from=jyrki@google.com;
 helo=mail-oi0-f41.google.com
Subject: Re: New Version Notification for
 draft-vkrasnov-h2-compression-dictionaries-01.txt
Archived-At: <http://www.w3.org/mid/CAPapA7TfTMdOA4XB=9b6959CUxqfZFe=A8nVnQB3xd3pG8HBRg@mail.gmail.com>
Resent-From: ietf-http-wg@w3.org
Resent-Sender: ietf-http-wg-request@w3.org
Precedence: list

--001a113eec1a3f8ac6054063126b
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

On Thu, Nov 3, 2016 at 6:43 AM, Vlad Krasnov <vlad@cloudflare.com> wrote:

> This is indeed not unlike SDCH and =E2=80=9Cquasi dictionaries=E2=80=9D o=
nly your
> dictionary defined by a stream and not a different url. In fact it can be
> used with SDCH just as well.
> My proposal tries to be as algorithm agnostic as possible.
> However brolti compresses much better in that case. In fact from what I
> have seen brotli+=E2=80=9Dquasi=E2=80=9D beats sdch+=E2=80=9Dquasi=E2=80=
=9D+brotli (but maybe
> sdch+=E2=80=9Dquasi=E2=80=9D+brotli+=E2=80=9Dquasi=E2=80=9D will do even =
better?).
>

Yes, this is consistent with our experiments. Brotli + custom dictionary is
more efficient than SDCH + custom dictionary with further gzip or brotli
compression.

With small (~150 kB) dictionaries brotli wins SDCH by 10-20 % (20 % at
brotli quality 11, 10 % at quality 6), with larger (1 MB) dictionaries
more, 20-30 % (30 % at brotli quality 11, 20 % at quality 6). The larger
the custom dictionary, the more powerful brotli comes in comparison to
SDCH. All our experiments indicate that SDCH is fully redundant and can be
abandoned if we add custom dictionary support to brotli content encoding.

--001a113eec1a3f8ac6054063126b
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div class=3D"gmail_extra"><div class=3D"gmail_quote">On T=
hu, Nov 3, 2016 at 6:43 AM, Vlad Krasnov <span dir=3D"ltr">&lt;<a href=3D"m=
ailto:vlad@cloudflare.com" target=3D"_blank">vlad@cloudflare.com</a>&gt;</s=
pan> wrote:<br><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px 0p=
x 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div style=
=3D"word-wrap:break-word"><div><div>This is indeed not unlike SDCH and =E2=
=80=9Cquasi dictionaries=E2=80=9D only your dictionary defined by a stream =
and not a different url. In fact it can be used with SDCH just as well.</di=
v><div>My proposal tries to be as algorithm agnostic as possible.</div><div=
>However brolti compresses much better in that case. In fact from what I ha=
ve seen brotli+=E2=80=9Dquasi=E2=80=9D beats sdch+=E2=80=9Dquasi=E2=80=9D+b=
rotli (but maybe sdch+=E2=80=9Dquasi=E2=80=9D+brotli+=E2=80=9Dquasi=E2=80=
=9D will do even better?).</div></div></div></blockquote><div><br></div><di=
v>Yes, this is consistent with our experiments. Brotli + custom dictionary =
is more efficient than SDCH + custom dictionary with further gzip or brotli=
 compression.</div><div><br></div><div>With small (~150 kB) dictionaries br=
otli wins SDCH by 10-20 % (20 % at brotli quality 11, 10 % at quality 6), w=
ith larger (1 MB) dictionaries more, 20-30 % (30 % at brotli quality 11, 20=
 % at quality 6). The larger the custom dictionary, the more powerful brotl=
i comes in comparison to SDCH. All our experiments indicate that SDCH is fu=
lly redundant and can be abandoned if we add custom dictionary support to b=
rotli content encoding.</div></div></div></div>

--001a113eec1a3f8ac6054063126b--