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

Re: SPDY Review

From: Mike Belshe <mike@belshe.com>
Date: Fri, 8 Jun 2012 15:55:42 -0700
Message-ID: <CABaLYCvRObLwoPZ18U4Typ7i0v-97d+b6aEFPZkz5JD4BekUTQ@mail.gmail.com>
To: Martin Nilsson <nilsson@opera.com>
Cc: Roberto Peon <grmocg@gmail.com>, ietf-http-wg@w3.org
On Fri, Jun 8, 2012 at 3:30 PM, Martin Nilsson <nilsson@opera.com> wrote:

> On Fri, 08 Jun 2012 21:10:59 +0200, Mike Belshe <mike@belshe.com> wrote:
>>>>  I'm using the same data as the http://www.eecis.udel.edu/~**
>>> amer/PEL/poc/pdf/SPDY-Fan.pdf<**http://www.eecis.udel.edu/~**
>>> amer/PEL/poc/pdf/SPDY-Fan.pdf<http://www.eecis.udel.edu/~amer/PEL/poc/pdf/SPDY-Fan.pdf>
>>> >**and when using the evaluation set they use I actually get somewhat
>>> better
>>> compression values (probably due to better header normalization).
>> The point of SPDY header compression is to use stateful compression.  If
>> you only compress one chunk of headers, you'll get modest savings -
>> 30-50%.
>>  but on the second set, you get like 95% :-)  For SPDY sessions in steady
>> state, the typical size of a request is 50-60 bytes.  Almost everything,
>> including the cookies, compresses out.
> Yes, that is quite clear from the specification, and I noted "deflate with
> persistent context is a good approach" in the original text. The context
> here is about using a deflate dictionary, and then the subsequent requests
> compress almost equally well without dictionary. This is natural given how
> LZ77 works and the repetitive nature of HTTP headers. So if you are only
> making single requests on every SPDY connection, the dictionary gives you a
> 15-25% boost on the header compression. If you are making multiple requests
> you will however very quickly see the benefit over normal deflate disappear.

OK.   Yes, we know that the dictionary only helps a small amount.  But it
does help.  The good news is that its basically free to use the dictionary,
so I don't really see any downsides.


> /Martin Nilsson
> --
> Using Opera's revolutionary email client: http://www.opera.com/mail/
Received on Friday, 8 June 2012 22:56:11 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:14:00 UTC