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

Re: FYI: Tools to evaluate header compression algorithms

From: Roberto Peon <fenix@google.com>
Date: Mon, 7 Jan 2013 11:02:36 -0800
Message-ID: <CAGzyod4cuRYE_uT7BM6+xRZ7N0TQ5E_eyOGAXvrZTW5tqnBsgg@mail.gmail.com>
To: Ilya Grigorik <ilya@igvita.com>
Cc: Roberto Peon <grmocg@gmail.com>, Martin J. Dürst <duerst@it.aoyama.ac.jp>, Mark Nottingham <mnot@mnot.net>, "ietf-http-wg@w3.org Group" <ietf-http-wg@w3.org>
As Mark already points out, this is pretty different as it only includes
the first page. Given how the compressors in general work, this is likely
to bias the results towards compressors that do better on only that page
load to the detriment of those which do better on further page loads.

We need traces of the "average user interaction" for each site to do a
really good job.

-=R


On Sun, Jan 6, 2013 at 12:34 PM, Ilya Grigorik <ilya@igvita.com> wrote:

>
>
> On Sun, Jan 6, 2013 at 1:55 AM, Roberto Peon <grmocg@gmail.com> wrote:
>
>> Do you have some suggestions Martin?
>> The obvious thing in my mind is to get submissions from site owners, but
>> that takes interest on their part first. :/
>>
>
> HTTP Archive is now scanning ~300K top domains (at least according to
> Alexa). While its still "top site" biased, I think that's a pretty good
> sample to work with. I believe we should be able to get the HAR files from
> it.
>
> ig
>
>
>
>>
>> On Sun, Jan 6, 2013 at 12:53 AM, "Martin J. Dürst" <
>> duerst@it.aoyama.ac.jp> wrote:
>>
>>> On 2013/01/06 14:57, Mark Nottingham wrote:
>>>
>>>> Quick follow-up:
>>>>
>>>> I posted more about this here:
>>>>    http://www.mnot.net/blog/2013/**01/04/http2_header_compression<http://www.mnot.net/blog/2013/01/04/http2_header_compression>
>>>>
>>>> In particular, we have graphs for all of the HAR samples I took earlier:
>>>>    http://http2.github.com/http_**samples/mnot/<http://http2.github.com/http_samples/mnot/>
>>>>
>>>
>>> These look very interesting. Just two points for the moment:
>>>
>>> - Drawing connected curves seems misleading, because we are not
>>> mesuring/showing a continuous quantity that varies over time, but discrete
>>> requests and responses.
>>>
>>> - The data sample includes big guys only. Some criticism of speedy has
>>> said that it is geared towards the big guys. Is there a way to get some
>>> more of an impression of how headers look at the long tail of websites?
>>>
>>> Regards,   Martin.
>>>
>>>
>>
>
Received on Monday, 7 January 2013 19:03:03 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 January 2013 19:03:05 GMT