Re: SPDY Header Frames

On Jul 17, 2012 2:15 AM, "Poul-Henning Kamp" <phk@phk.freebsd.dk> wrote:
>
> In message <CABP7Rbfg6nf=
qJOSAyC8mGcfDQxFmxREOqmb88_fSqiyc5Kv+Q@mail.gmail.com>
> , James M Snell writes:
>
> >Yeah.. um.. ignore that compression ratio, lol...
>
> I would have checked it but I have become allergic to dissecting
> gzip files after implementing ESI:includes on gzip'ed files in
> Varnish :-)
>
> >[...]so if our header encoding is compact enough without it, why spend
> >the additional cycles to save 20 bytes or so.
>
> Indeed.
>
> Speaking of which:  I was surprised to find that the SPDY dictionary
> didn't contain any fodder for compressing User-Agent ?
>
> Was this future-proofing/neutrality, to avoid trademark issues or
> because there were no reasonable general benefit to be had ?

A combination of future proofing and neutrality, as you've guessed.

We didn't want to provide any motivation for people to lie about user-agent
any more than they do today, and having something in the dictionary would
have provided such a motivation.
We also didn't want the protocol to "choose" any favorites, i.e. we wished
it to be neutral.
-=R

>
> (And what DO we do with User-Agent in HTTP/2.0, it must be possible
> to come up with something actually usable... ?
>
> >I am interested in hearing what further optimizations could be made to
the
> >structure.
>
> I take it that you have read draft-tarreau-httpbis-network-friendly-00 ?
>
> It has a lot of good ideas...
>
> --
> Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
> phk@FreeBSD.ORG         | TCP/IP since RFC 956
> FreeBSD committer       | BSD since 4.3-tahoe
> Never attribute to malice what can adequately be explained by
incompetence.
>
 On Jul 17, 2012 2:15 AM, "Poul-Henning Kamp" <phk@phk.freebsd.dk> wrote:

> In message <CABP7Rbfg6nf=
> qJOSAyC8mGcfDQxFmxREOqmb88_fSqiyc5Kv+Q@mail.gmail.com>
> , James M Snell writes:
>
> >Yeah.. um.. ignore that compression ratio, lol...
>
> I would have checked it but I have become allergic to dissecting
> gzip files after implementing ESI:includes on gzip'ed files in
> Varnish :-)
>
> >[...]so if our header encoding is compact enough without it, why spend
> >the additional cycles to save 20 bytes or so.
>
> Indeed.
>
> Speaking of which:  I was surprised to find that the SPDY dictionary
> didn't contain any fodder for compressing User-Agent ?
>
> Was this future-proofing/neutrality, to avoid trademark issues or
> because there were no reasonable general benefit to be had ?
>
> (And what DO we do with User-Agent in HTTP/2.0, it must be possible
> to come up with something actually usable... ?
>
> >I am interested in hearing what further optimizations could be made to the
> >structure.
>
> I take it that you have read draft-tarreau-httpbis-network-friendly-00 ?
>
> It has a lot of good ideas...
>
> --
> Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
> phk@FreeBSD.ORG         | TCP/IP since RFC 956
> FreeBSD committer       | BSD since 4.3-tahoe
> Never attribute to malice what can adequately be explained by incompetence.
>
>