Re: [hybi] workability (or otherwise) of HTTP upgrade

[Adam Barth <[e-mail]>]

On Mon, Dec 6, 2010 at 11:29 PM, Willy Tarreau <[e-mail]> wrote:
> On Tue, Dec 07, 2010 at 05:14:02PM +1100, Mark Nottingham wrote:
>> On 07/12/2010, at 5:10 PM, Willy Tarreau wrote:
>>
>> > In my opinion the problem is not here, but the adoption rate depending
>> > on the port. Many organisations implement URL filtering on port 80,
>> > white-list based filtering on 443 and nothing else around. If you want
>> > to deploy a site which quickly gets a lot of traffic, port 80 clearly
>> > is the most suited, which is even more true considering that long polling
>> > mechanisms already work over that port.
>>
>> Quantify 'many.' According to Adam's paper, ~13% of clients will fail to negotiate with a CONNECT-based solution. Is "many" > 13% of the Internet?
>
> We have no idea where it was measured.

It was measured on the Internet.  We placed an advertisement on an ad
network containing the experiment.  The ad was shown globally,
according to the audience of the ad network.

> If it's in a campus, 13% of failure is
> high, considering the general low amount of filtering at those places. If it
> was in corporate networks, 13% is low.

The experiment was not designed to measure success rates, so I
wouldn't read too much into them.

Adam