W3C home > Mailing lists > Public > ietf-http-wg@w3.org > October to December 2012

Re: SPDY and the HTTP Binding

From: Roberto Peon <grmocg@gmail.com>
Date: Fri, 12 Oct 2012 14:49:20 -0700
Message-ID: <CAP+FsNfzz7Wu9bQRcYHeZ2gNA9s5HOY0fcFVqnhvbqD14LiKHw@mail.gmail.com>
To: Willy Tarreau <w@1wt.eu>
Cc: Poul-Henning Kamp <phk@phk.freebsd.dk>, James M Snell <jasnell@gmail.com>, ietf-http-wg@w3.org
The most recent output, copy/pasted is:

"Delta-coding took: 0.642199 seconds for: 104300 header frames or
6.15723e-06 per header or 162411 headers/sec or 8.88429e+07 bytes/sec"

So, ~89 million bytes/second and 162k requests/second for the delta-coding
on one core. Given that most machines have more than once core, this
performance seems pretty reasonable to me, especially given that it is not
really been optimized. I'd hope that bandwidth was predominantly
entity-bodies and not headers, and so the bandwidth doesn't seem
problematic to me either.

The input data for this was a 1000 iterations over the following har files
representing full page loads:

mail.google.com.har pinterest.com-bicycle-search.har
pinterest.com-homepage.har www.amazon.com-bicycle-search.har
www.amazon.com-homepage.har www.bing.com-bicycle-search.har
www.bing.com.har www.cnn.com-homepage.har www.facebook.com-homepage.har
www.google.com-bicycle-image-search.har www.google.com-bicycle-search.har
www.google.com-homepage.har

I'd prefer to have more comprehensive test data, but this is a decent start.

-=R



On Fri, Oct 12, 2012 at 1:47 AM, Willy Tarreau <w@1wt.eu> wrote:

> On Fri, Oct 12, 2012 at 08:07:23AM +0000, Poul-Henning Kamp wrote:
> > --------
> > In message <CAP+FsNcp7r2ad-X=2aNYq=_
> FB8cKgKEpTUYW7eG7YsyFnrwbJg@mail.gmail.com>, Roberto Peon write
> > s:
> >
> > >In terms of HTTP requests/second I think it is on the
> > >order of 30k/core.
> >
> > This is why I think we need to do an envelope+content like format:
> > HTTP-routers will need to do several orders of magnitude faster
> > than that in a few years.
>
> Not only in a few years, but already today. I have some haproxy
> users who run it at 100kreq/s already. That's why I'm extremely
> picky about the CPU impact of compression.
>
> Willy
>
>
Received on Friday, 12 October 2012 21:49:48 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 12 October 2012 21:49:51 GMT