[hybi] WebSocket over HTTP2 is discussed on httpbis

[Yutaka Hirano <[e-mail]>]

Hi,

As I announced before, WebSocket over HTTP2 is discussed on httpbis.
Though tunneling is widely supported in HiBy, httpbis people don't like it
and I failed to persuade them.
Currently there are two proposal [1] and [2].
[2] maps a WebSocket message to HEADERS + DATAs HTTP2 frames and it doesn't
preserve RFC6455 semantics.
For Tobias and Peter, I would show how [2] meets your design priorities
given before.

Tobias's priority:

> 1. allows to provide W3C WebSocket API
>
Yes.

> 2. no "80%-WebSocket-protocol-compatible" design (either compatible with
> full RFC6455 incl. any extensions or outright independent of WS)
>
Yes, 0% RFC6455 Frame compatible (though I would like to  reuse some
handshake stuff).

> 3. does not put servers that provide both native WS and SPDY (but not any
> SPDY/WS) on same origin at a disadvantage
>
It doesn't depend on the framing problem.

> 4. minimizes code coupling between WS and SPDY within implementations

No, though the mapping is simple.

Peter's priority:

> Important:
> 1. W3C WebSocket API and muxing go without saying.
>
Yes.

> 2. Dedicated RFC6455 connections remain a viable option in modern browsers.
>
Yes.

> 3. Efficient and clearly defined method for detecting HTTP/2.0 messaging
> support, choosing HTTP/2.0 vs raw RFC6455, when to fall back, etc.
>
It doesn't depend on the framing stuff.

> 4. Payload compression with appropriate configurability (either via
> permessage-deflate support or a native mechanism)
>
it is possible.

Nice to haves:
> 5. Avoid significant regressions from RFC6455 in wire overhead, latency,
> and per message fixed costs for very small (<100byte) messages.
>
I think there is no significant regressions.

> 6. Clean spec/avoiding redundant and unnecessary features of RFC6455.
>
Yes.

> Only if it doesn’t conflict with 1-6:
> 7. Options for future compatibility with simple RFC6455 extensions

No.


And, protocol negotiation should be also discussed.
In fact, it has more impact on the HTTP2 spec and we should provide
feedback to them as soon as possible if any (See [3] for example).
Please join discussions on http2bis if you have any comments.

Thanks,

[1]:
https://github.com/yutakahirano/ws-over-http2/blob/master/draft-hirano-websocket-over-http2.txt
[2]:
https://github.com/yutakahirano/ws-over-http2/blob/master/ws-over-http2-message-mapping.md
[3]: http://tools.ietf.org/html/draft-nottingham-httpbis-alt-svc-03