Re: [TLS] Bakeoffs

[Michael Sweet <[e-mail]>]

Patrick,

On Apr 17, 2014, at 9:41 AM, Patrick McManus <[e-mail]> wrote:
> 
> On Thu, Apr 17, 2014 at 8:19 AM, Michael Sweet <[e-mail]> wrote:
> >> FWIW, 1-RTT latency is mainly an issue when you have a browser making a dozen connections to a secure site.  With HTTP/2 that particular problem goes away...
> 
> As a browser guy I'll say I really disagree with that.

As a printer guy, I can definitely say that browsers opening dozens of connections is a major source of latency, since the printer's ability to respond to so many simultaneous connections is limited.  It's even worse when you tunnel connections over USB (IPP USB protocol) and have to multiplex requests over a limited number of channels (2 in most cases).

I'll grant that many web sites pull content from multiple servers/domains, in particular to support advertising, however it is uncommon (at least in my experience) to find all of a page's resources spread to different servers.

> HTTP/2 is important stuff, and helps hide part of the problem, but the latency issue continues to be an issue for HTTP over TLS use cases. Time to first byte of course, but also to each different referenced origin (the cdn, the third-party ad network, the identity provider, etc..). Often things that block layout are on that cdn (e.g. stylesheets, some javascript) so the time to perform 2 http transactions on 2 serialized tls terminated connections is often the same as time to first screen draw. Its considerably worse in mobile environments where RTT can be out of control.

Indeed, especially when you are opening dozens of connections before even doing a TLS handshake.

> Establishing TLS is noticeably slow, and that's a problem for the browser use case where https often competes against http for developer mind share. As a natural result, some content providers choose to run plaintext http which we can all agree is an undesirable outcome. Shrinking that gap will help.

Based on the push back in the HTTP/2 discussions on that subject, I don't think you'll get unanimous agreement on that. TLS has its place, and it is important for us to have a solid protocol to support confidential communications, but there are significant costs associated with TLS - performance, maintenance, power, money, etc. - that prevent it from being adopted wholesale today.

_________________________________________________________
Michael Sweet, Senior Printing System Engineer, PWG Chair