Re: [rtcweb] Data Channel Protocol: Data before DATA_CHANNEL_ACK

[Randell Jesup <[e-mail]>]

On 2/10/2014 10:20 AM, Peter Thatcher wrote:
> If you have to wait for the ACK, it's one more RTT you have to wait 
> before you can send data.  We've already got many RTTs form ICE and 
> DTLS and what's built into SCTP.  We originally didn't have an ACK 
> message at all, in large part because we didn't want more RTTs.  The 
> ACK was put in to handle the edge case of unordered data arriving 
> before the OPEN message.  But we added it in a way that it doesn't add 
> more RTTs.

Exactly.   Especially at call-start.

>
>
> On Mon, Feb 10, 2014 at 2:20 AM, Christer Holmberg 
> <[e-mail] 
> <mailto:[e-mail]>> wrote:
>
>     Hi,
>
>     As is defined in the data channel protocol, an entity can send
>     data once DATA_CHANNEL_OPEN has been sent, before the associated
>     DATA_CHANNEL_ACK is received.
>
>     What is the reason for allowing data to be sent before
>     DATA_CHANNEL_ACK? The sender may not even know (depends on
>     whatever external negotiation mechanisms are used) whether the
>     remote peer supports the protocol to begin with.
>
>     It think it would be good to allow the remote peer to accept
>     (DATA_CHANNEL_ACK) or reject (stream reset) the DATA_CHANNEL_OPEN
>     before data is sent.
>

"Rejection" of a channel can be done be either ignoring it (drop the 
channel reference on the floor), or actively closing it 
(channel.close(), then drop it on the floor).  This was agreed on a long 
time ago.

-- 
Randell Jesup
[e-mail]