Re: [TLS] Next protocol negotiation

[Adam Langley <[e-mail]>]

On Thu, Jan 21, 2010 at 12:07 AM, Yoav Nir <[e-mail]> wrote:
> That HTTPS number is very surprising. It suggests that 5% of Chrome users can do regular HTTPS, but not WebSockets over HTTPS?
>
> Are SSL-inspecting proxies that prevalent?
>
> Maybe this suggests that Chrome's bias is FOR tightly-controlled environments rather than AGAINST.

As noted, this data is preliminary but I will make some guesses:

The nature of the test is that it proceeds in steps, each step testing
some behaviour of the WebSockets protocol. If the user closes the
browser during the process, this can appear the same as a protocol
failure.

The breakdown of the failure rates at each step with TLS looks very
different from the HTTP patterns. With HTTP we see clear jumps where
the network is opaque to some behaviour for a significant fraction of
our users. The TLS data increases nearly linearly in comparison. Thus
it might well be that we're just seeing noise in the TLS data as the
steps proceed over time. I don't believe that TLS intercepting proxies
are actually deployed in front of 5% of our users.

We obviously want to investigate this ourselves and we're continuing
with our analysis and data gathering.


AGL