Re: [rtcweb] Support for websockets

[Christopher Blizzard <[e-mail]>]

On 9/4/2011 8:12 PM, Matthew Kaufman wrote:
>
> I think there's a legitimate question as to how transmission of media 
> over TCP should work. I believe that the existing code bases all 
> assume that if you get IP addresses, you use them; if you get IP 
> addresses for TURN servers you talk to them over UDP and use TURN for 
> relaying; and if you get IP addresses for TURNS servers you talk to 
> them over TLS/TCP and use them for relaying. Therefore the only TCP 
> transport is TURN-over-TLS-over-TCP.
>
> Is that sufficient and reasonable, or should media-over-websockets (or 
> something else) be how TCP transmission of media works?

It's important to note that WebSockets aren't raw sockets in the classic 
TCP or POSIX sense.  So a conversation about transmission of media over 
TCP doesn't really apply to WebSockets, exactly.  It's true that since 
WS is over TCP that it's reliable and ordered.  What would be 
interesting would be a discussion of how to take media data coming in 
over a WebSocket and feed it to a consumer that could display that 
media, as well as the reverse.  But an API discussion feels like 
something that's more that's something that belongs at the W3C.

If we wanted a standardized representation in WS that might happen here, 
but it's not something that's strictly required.  Well-built APIs to 
encoders and decoders could mean that it's up to page implementers to 
figure out how to package the data, manipulated by JS.

--Chris