Why HTTP/2.0 does not seem interesting (2012)

https://www.varnish-cache.org/docs/trunk/phk/http20.html?re

by asm89 • 13 years ago

189 47 13 years ago

A good analysis but somewhat dated. It has been established, for instance, that with a compatible SSL lib and TLS NPN support SPDY/HTTPbis can be supported on the same socket and port as HTTPS. With that said I really like what PHK has to say about "HTTP routers".

mortehu• 13 years ago

His suggestion is:

One simple way to gain a lot of benefit for little cost in this area, would be to assign "flow-labels" which each are restricted to one particular Host: header, allowing HTTP routers to only examine the first request on each flow.

I don't understand what he's saying here. Who assigns a label to the "Host" header when? Is he proposing sticky cookies?

exceptione3• 13 years ago

  In my view, HTTP/2.0 should kill Cookies as a concept,
  and replace it with a session/identity facility, which makes 
  it easier to do things right with HTTP/2.0 than with HTTP/1.1.

count me in. Cookies are a huge waste of bandwidth and freaking annoying here in Europe as you cannot visit a site anymore without being warned you are about to receive yet even more cookies.

yahelc• 13 years ago

Cookies are...freaking annoying here in Europe as you cannot visit a site anymore without being warned you are about to receive yet even more cookies.

Seems like the blame for that lies not with cookies themselves, but with the EU's cookie law.

exceptione3• 13 years ago

Well the law is implementend in an annoying fashion. For functional cookies there is no requirement to ask permisson. But many websites choose to block all content asking to permit them to place additional cookies for benefits, i.e. tracking cookies.

Like the author said:

  Cookies are, as the EU commision correctly noted, 
  fundamentally flawed, because they store potentially 
  sensitive information on whatever computer the user 
  happens  to use, and as a result of various abuses and
  incompetences, EU felt compelled to legislate a "notice 
  and announce" policy for HTTP-cookies.

Bosence• 13 years ago

Oh yes, the regulation was rushed in by people not really sure what to do or what was happening.

Why it couldn't be a browser option, I have no idea.

drostie• 13 years ago

At best that's a symptom.

Cookies are for session management; the central problem with cookies is that people feel that servers will treat certain sessions as ephemeral, but instead those servers track these people for a long-term creepy analysis. One connected problem is that many sites require cookies in order to show public content. Public-content sessions should be entirely ephemeral, meaning that you shouldn't need a cookie in the first place. (The New York Times offends in this regard egregiously and persistently.)

You can easily comply with the EU law by either placing the notice on the login page or else not storing cookies. This means that anybody who abuses cookies in the above way needs to be loud about it; "we're not giving you an ephemeral presence like you think!" -- which actually not only fixes this problem but also creates an incentive to not abuse cookies in this way.

I am not saying that we should abandon sessions entirely, but that it would be nice if the 'default' session treatment followed the rules that online banking uses: when the browser is closed, all sessions are done. If we did this then we'd want to include a 'persistent login' mechanism, which would take the form of an in-browser 'would you like to sign in?' dialogue accompanying a web site. This means that unlike current HTTP authentication, it would have to be somewhat asynchronous; you are shown the ephemeral version of the page while the browser itself requests you to confirm that you want to join your long-term session there. (I was originally going to recommend that the browser just handle a digital signature scheme, but of course that does not solve the 'logging on to Facebook from your sister's computer' problem easily. Hm.)

andyhmltn2• 13 years ago

Please correct me if I'm wrong, but wasn't that the UK's law? If it was, they seemed to backtrack on it and say that it's fine as long as you mention it in your site's TOS

yalogin2• 13 years ago

What is the alternative to cookies? What does he mean by session/identity facility?

_delirium2• 13 years ago

There's some discussion in this thread: http://lists.w3.org/Archives/Public/ietf-http-wg/2012JulSep/...

in short,

   the need for a Session header to replace the use of Cookies
   for basic session management

solox3• 13 years ago

The EU "Cookie law" is not limited to cookies. The term used in regulation 6 is "storage", which includes at least localStorage, headers, and remotely hosted solutions (off the top of my head).

DrJokepu• 13 years ago

Basically, I can guarantee that killing cookies would lead to 0% adoption of HTTP 2.0 forever, due to the same inertia that is holding Python 3 or IPv6 back. Doesn't matter if there's a better mechanism included that does the same job.

exceptione3• 13 years ago

…unless the EU forbids tracking and session cookies all together, but a ban on tracking cookies alone will be enough motivation already i guess.

takeda64• 13 years ago

Provide session support in HTTP and still cookies for backward compatibility. If the support works right, people will migrate themselves.

jessaustin3• 13 years ago

I always thought that rfc2616 felt more "pure" since it didn't mention cookies. I mean, http was supposed to be stateless right?

_delirium2• 13 years ago

Some discussion from last year, fwiw: https://news.ycombinator.com/item?id=4253538

richardwhiuk• 13 years ago

What really frustrates me about HTTP as a protocol is that it provides the beginning of a framework to do session management using the WWW-Authenticate headers, but it's ignored because the site can't provide a good UX. Instead we end up with phishing, terrible login forms and poor security when people reimplement session management in Cookies.

jessaustin3• 13 years ago

I've long wondered if we might have had better alternatives via WWW-Authenticate if major browsers had made it straightforward/possible to write auth plugins. (Actually it probably is possible, but AFAICT not without non-portable munging about in NPAPI.) If Mozilla actually do something to integrate Persona into their clients, will they do so in an open, repeatable way (with an API accessible to extensions) or will it just be more of the same oddball one-off coding that supported NTLM?

prima-facie• 13 years ago

This a thousand times. Why didn't they fix this already?

osth• 13 years ago

I wonder if header compression is primarily to allow for ubiquitous, large cookies.

Cookies were originally and with few exceptions remain a hack to try to add state to transactions that were not intended to be stateful.

If indeed the header compression is driven by the growing prevalence and size of cookies, then HTTP/2 is an effort to accomodate a hack. Not very interesting.

Some hacks that find their way into RFC's are difficult to remove because the transition process would be unreasonably expensive, like replacing the "sophomoric" compression scheme in DNS with something more sensible like LZ77 (credit: djb). I guess we might see some passionate arguments by web developers about the great expense of removing cookies from the HTTP standard and replacing it with a session facility, but I think the (long term) benefits easily outweigh the (short term) costs.

jokoon• 13 years ago

Well transmitting text never seemed really very much interesting.

http was nice because it was easy for software programmers to write apps that could work over http, because no binary protocol was involved: reading ASCII strings is never complicated. It was good for a growing industry.

Now most browsers are open source, why can't the IETF work out a binary protocol ? Bittorrent is binary, and it's awesome and it's used. Why can't any browser work out a binary protocol ? Truly dynamic pages over the network ? Why not ? I'm sure some software already does that. Make one open source, make it work on firefox or chrome, I guess things would start to light up.

cromwellian• 13 years ago

This has been going on for a long time and nothing got done. HTTP-NG started in 1997. If you try to boil oceans, you get nothing.

The recent why SPDY is succeeding is because it is not trying to solve every problem and grind every axe.

tracker1• 13 years ago

In reading this a protocol that supports two initial statements upon connection/negotiation as follows...

    s: http/2.0 {SERVER INFO}
    c: connect host/   <-- no path
    s: OK {server-cert/key}
    -- all futher requests encrypted against public key/cert
    c: session-start {client key/cert}
    s: SESSION: {session id} ({domain1},{domain2},...)
    c: (COMMAND|get|put|post|delete) {PATH}
    s: OK
       or
       DENIED ### (reason)   <-- response code & reason
       or
       REDIRECT host/(path)  <-- if the file is physically on another backend
    c: {OTHER REQUEST HEADERS START}

after a session is started, the client may make other requests

    s: http/2.0 {SERVER INFO}
    c: connect host/{path}
    s: OK {server-cert/key} or DENIED ### Reason
    -- all futher requests encrypted against public key/cert
    c: session-join {SESSION_ID} {client key/cert}
    s: OK or DENIED...
    c: {COMMAND} {path}

from there, the "session_id" can be a key for server-side value storage/lookup, etc... sent over the encrypted channel

mbetter• 13 years ago

Our general policy is to only add protocols if we can do a better job than the alternative, which is why we have not implemented HTTPS for instance.

asm89• 13 years ago

Longer explanation of why there is no SSL in Varnish can be found here: https://www.varnish-cache.org/docs/trunk/phk/ssl.html.

Why HTTP/2.0 does not seem interesting (2012)

Related Stories