W3C home > Mailing lists > Public > ietf-http-wg@w3.org > January to March 2013

Re: Framing and control-frame continuations

From: Roberto Peon <grmocg@gmail.com>
Date: Wed, 6 Feb 2013 06:06:11 -0800
Message-ID: <CAP+FsNf1s3CV5eMzfmhFa2C8qf04uef732s690odE3uTcNTkGQ@mail.gmail.com>
To: Poul-Henning Kamp <phk@phk.freebsd.dk>
Cc: Mark Nottingham <mnot@mnot.net>, HTTP Working Group <ietf-http-wg@w3.org>
I know from past experience that one can fill up a 10GE pipe with very
small SPDY frames (1XXmillion 1 byte payloads)  when done correctly on
today's hardware, without using up all cores (though, to be fair it comes
close). This is pretty much a worst-case scenario.
I've found that the amount of cores one uses to output a certain amount of
bandwidth is far more dependent on the the NIC kernel drivers and NIC
hardware than anything else. Formatting things in memory *especially* data
frames is extremely cheap given that it requires no transformation.

I currently can't imagine what a 1T network architecture will look like
when going into a single machine.  I don't know if it'd go through the
kernel, or just DMA somewhere, or how programmable the DMA controllers will
be, etc.
I do know, though, that I could cram ~over 100M SPDY frames over the wire
on a 10GE interface on today's hardware, so I'm not worried about shoving
16M 64k frames over the wire in terms of overhead.
I do also know is that a terrabyte could be sent ~16M 64k frames. Even
assuming we do terribly and only stuff the frames with 16k, we'd only have
a piddly 67M frames per second to deal with. That is already lower than
what we can accomplish today.

-=R



On Wed, Feb 6, 2013 at 3:40 AM, Poul-Henning Kamp <phk@phk.freebsd.dk>wrote:

> Content-Type: text/plain; charset=ISO-8859-1
> --------
> In message <C6B30EDB-32FB-4093-AD74-4B596FC4F0D2@mnot.net>, Mark
> Nottingham wri
> tes:
>
> >> I don't think this is good enough.
> >>
> >> You'd need ~20k of these frames a second to fill a 10GB ethernet,
> >> and a very large fraction of present day web-objects would require
> >> more than one frame already.
> >
> >Could you please spell out why you think that's a problem?
>
> As I said before:  If we cannot demonstrate HTTP/2 on a 10GE
> media and argue plausibly that it can be handled on 1Tbit/sec,
> we're writing a standard for the past and not the future.
>
> Matt Mathis has been pushing similar agenda for IP MTU for years now:
>
>         http://staff.psc.edu/mathis/MTU/
>
> --
> 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.
>
Received on Wednesday, 6 February 2013 14:06:40 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Wednesday, 6 February 2013 14:06:42 GMT