Working Group assembling to discuss HTTP/2.0 | Hacker News Reader

Working Group assembling to discuss HTTP/2.0

lists.w3.org

80 pointsmetabrew14 comments

While it would be exciting to see something as progressive as SPDY become codified as a standard, keep in mind the deliberately conservative nature of standards development, as seen in the responses to that message:

  On 24/01/2012, at 3:50 PM, James Snell wrote:
  > +1... would love to see that work move forward within this group. I do
  > have concerns about keeping the scope focused, however. Even tho the
  > door would be opened to the possibility of new features being
  > introduced, there is obvious danger in opening those doors too wide. I
  > strongly feel that things such as the introduction of new request
  > methods and new header fields, unless there is clear and irrefutable
  > evidence of their general utility within the core of the spec, should
  > continue to be pursued as they are today -- within separate I-D's as
  > extensions to the core protocol. The charter should make it absolutely
  > clear that the goal is an incremental evolution of HTTP/1.1 rather
  > than an opportunity for radical changes.

  Thanks, James.
  Pretty much everyone I've spoken to about this has raised the same concern.
  I agree it's going to be a tightrope walk, but I think there's a healthy 
  amount of concern about this in the community, which will help guide development.
  Cheers,
  Mark Nottingham

Whilst SPDY is nice, the compression possible using SDCH is a game changer. Given the amount of disk space on machines it should be possible to send quite large dictionaries to browsers.

For example, I've been experimenting with the Bentley/McIlroy compression algorithm and I can easily (no effort at all) compress the BBC News web site by 90% (i.e. to 10% of its original size).

For some reason SDCH gets no respect. Somebody could become a hero by writing an auto-SDCHing reverse proxy.

The stuff that's big (rich media) already has its own native compression. The stuff that's relatively small (text) also already compresses pretty well without shared-dictonaries.

I think that's the reason: it's not that big a win over the total traffic of a modern browsing session. For example, I doubt jgrahamc's experiment yields a 90% savings over the entire browsing session with BBC News, when getting images, and if gzip is already enabled on the HTML/CSS/JS. I'd guess closer to 10%.

Updated after taking a quick look at BBC News homepage in Firebug's net panel:

The BBC News homepage has about 150KB of HTML/CSS… but they haven't even bothered to turn on gzip, which might bring that down to 30KB. It has about 230KB of images, which wouldn't be helped by a shared dictionary.

If they turned on gzip, text would be about 30/230=13% of the bandwidth. Even 99.99% compression – a dictionary of exactly today's content already on the client! – could only shrink the homepage download by 13%.

My impression from working on Chrome (but not near the network bits) is that while SDCH does have the potential for compression you mention, doing so requires tuning beyond what most organizations can do. And more importantly, total download size isn't as much in the critical path for performance as the latency and pipelining tricks SDPY does.

You can see a recent mail from the SDCH master on this subject here:

https://groups.google.com/group/SDCH/browse_thread/thread/7d...

I'm looking forward to the inevitable discussions about cross-origin requests, same-origin policy and all that jazz.

Having implemented SPDY/2, they could do a lot worse than just ratifying SPDY (v3 by now..) and perhaps clarify some of the CORS stuff.

I love the idea of SPDY, but I would really like to see an open source (transparent) proxy, then I could proxy all my internet traffic through the cloud.

Not quite what you want, but a good DefCon talk with code about making a 'real' session layer.

http://www.youtube.com/watch?v=aakzkrl-34g

Not sure why everybody is so ecstatic about Spdy... it has a bunch of warts, like for instance the shared dictionary for HTTP headers that saves a couple hundred bytes over connections that may send several MiB (pointless compression, complexity). Also the whole concept of priorities, cancelled requests still filling up network buffers, etc.

A committee of experts is how a new protocol like this should be developed, and I fully expect whatever they come up with to be better designed and considered than Spdy.

Why is it that every mailing list I read about updates to legacy design (even classic ones that have been successfully implemented and no one would change back) always that that guy (or guys) who say "no this is stupid, make your own widget if you want new stuff, floppy disks are a good enough standard"? Poul-Henning Kamp is that guy here.

The guy who coined the term "bikeshed color" is now arguing over the name of the next iteration of HTTP, and how since it can't be done in a year it shouldn't be done at all. I understand the need for conservatism, but outright pooh-poohing is never effective without major concerns.

I'd like HTTP over UDP for javascript applications.

Can you elaborate on how UDP would benefit JS? Have you seen Google's TCP fast start proposal (from yesterday, I believe)?

PeerConnections (will soon) give you pseuedo-UDP-socket-like ability in JavaScript. That with Media Source APIs and device access, one can build a p2p skype clone exclusively in the browser.

Frankly, your sentence doesn't make much sense. HTTP is a way of sending text and data, so UDP would never be used. The data needs to arrive, be in order, and be correct (exactly what TCP does). Also, the fact that a new HTTP protocol is being used has no effect on javascript, since the protocol is essentially abstracted away.

As for your real question, when will javascript have UDP socket support, that is an interesting question. The feature would definitely be handy for multiplayer games and video/audio streaming. In light of this, I'm sure something will come down the pipe eventually. There could be security worries though, for example it could be used to build a webpage that performs a DDoS (not that they can't be built already).

Reply on news.ycombinator.com