W3C home > Mailing lists > Public > ietf-http-wg@w3.org > January to March 2010

RE: httpbis and deflate compression...

From: Mahdavi, Jamshid <jamshid.mahdavi@bluecoat.com>
Date: Thu, 25 Mar 2010 18:12:51 -0700
Message-ID: <B0D803D04AA2F54A8227E1B606344ED9032064F9@bcs-mail04.internal.cacheflow.com>
To: "David Morris" <dwm@xpasc.com>, "Julian Reschke" <julian.reschke@gmx.de>
Cc: "Dan Winship" <dan.winship@gmail.com>, <ietf-http-wg@w3.org>

That is consistent with what we've seen / heard; deflate never or almost never actually gets used because gzip is usually also an option and tends to be preferred.

--J

-----Original Message-----
From: David Morris [mailto:dwm@xpasc.com]
Sent: Thu 3/25/2010 6:08 PM
To: Julian Reschke
Cc: Dan Winship; Mahdavi, Jamshid; ietf-http-wg@w3.org
Subject: Re: httpbis and deflate compression...
 

I have a limited sample of wild web proxy data streams which I could
check for what is used.

In a data sample that lasted about 5 minutes, I captured 4500+ 
request/responses. Of these:
   648 has a non-empty Content-encoding
   638  gzip
     2  none
     1  UTF-8
     6  sdch,gzip
In the same data
  2436 have a non-empty accept encoding on the request
  2433 were gzip, deflate
     3 were gzip

There were 836 unique host names in the data. (x.google.com and 
y.google.com are unique). And roughly 350 unique sites.

In this sample, as processed, there were no deflate encoded responses.

If someone could provide sniff like rules I could apply to deflate
content, if I find any, to determine which variation it is, I can
examine the data again as well as some additional data I have to
see if deflate is used and how many sites use it. 

This user population is probably biased to variations of IE but
I don't have user agent analysis handy.

Ideally, you'd send a python regex I could just fold into my tool.

Dave Morris

On Tue, 23 Mar 2010, Julian Reschke wrote:

> On 23.03.2010 02:48, Dan Winship wrote:
> > On 03/22/2010 08:49 PM, Mahdavi, Jamshid wrote:
> > > We could try to clarify this by adding an implementation note to the
> > > text. Alternately, since some servers actually do implement *just* the
> > > 1951 version, we could add another content-coding which is
> > > specifically for this.
> > 
> > My understanding is that the "bare" version is actually more common on
> > the web than the correct version, due to the fact that at least IE6 and
> > IE7 (not sure about 8 and 9) only support the bare version, and most
> > other browsers support both.
> 
> Sounds like it would be cool to have test cases...
> 
> > ...
> 
> BR, Julian
> 
Received on Friday, 26 March 2010 01:14:36 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 27 April 2012 06:51:17 GMT